LICENSE
MANIFEST.in
README.md
pyproject.toml
setup.py
miles/__init__.py
miles/backends/__init__.py
miles/backends/experimental/__init__.py
miles/backends/experimental/fsdp_utils/__init__.py
miles/backends/experimental/fsdp_utils/actor.py
miles/backends/experimental/fsdp_utils/arguments.py
miles/backends/experimental/fsdp_utils/checkpoint.py
miles/backends/experimental/fsdp_utils/lr_scheduler.py
miles/backends/experimental/fsdp_utils/parallel.py
miles/backends/experimental/fsdp_utils/update_weight_utils.py
miles/backends/experimental/fsdp_utils/kernels/__init__.py
miles/backends/experimental/fsdp_utils/kernels/fused_experts.py
miles/backends/experimental/fsdp_utils/kernels/fused_moe_triton_backward_kernels.py
miles/backends/experimental/fsdp_utils/models/__init__.py
miles/backends/experimental/fsdp_utils/models/qwen3_moe.py
miles/backends/experimental/fsdp_utils/models/qwen3_moe_hf.py
miles/backends/megatron_utils/__init__.py
miles/backends/megatron_utils/actor.py
miles/backends/megatron_utils/arguments.py
miles/backends/megatron_utils/bridge_lora_helpers.py
miles/backends/megatron_utils/checkpoint.py
miles/backends/megatron_utils/ci_utils.py
miles/backends/megatron_utils/fp32_param_utils.py
miles/backends/megatron_utils/initialize.py
miles/backends/megatron_utils/lora_utils.py
miles/backends/megatron_utils/misc_utils.py
miles/backends/megatron_utils/model.py
miles/backends/megatron_utils/model_provider.py
miles/backends/megatron_utils/parallel.py
miles/backends/megatron_utils/replay_utils.py
miles/backends/megatron_utils/sglang.py
miles/backends/megatron_utils/megatron_to_hf/__init__.py
miles/backends/megatron_utils/megatron_to_hf/deepseekv3.py
miles/backends/megatron_utils/megatron_to_hf/glm4.py
miles/backends/megatron_utils/megatron_to_hf/glm4moe.py
miles/backends/megatron_utils/megatron_to_hf/llama.py
miles/backends/megatron_utils/megatron_to_hf/mimo.py
miles/backends/megatron_utils/megatron_to_hf/qwen2.py
miles/backends/megatron_utils/megatron_to_hf/qwen3_5.py
miles/backends/megatron_utils/megatron_to_hf/qwen3_next.py
miles/backends/megatron_utils/megatron_to_hf/qwen3moe.py
miles/backends/megatron_utils/megatron_to_hf/processors/__init__.py
miles/backends/megatron_utils/megatron_to_hf/processors/padding_remover.py
miles/backends/megatron_utils/megatron_to_hf/processors/quantizer_compressed_tensors.py
miles/backends/megatron_utils/megatron_to_hf/processors/quantizer_fp8.py
miles/backends/megatron_utils/megatron_to_hf/processors/quantizer_mxfp8.py
miles/backends/megatron_utils/megatron_to_hf/processors/quantizer_nvfp4.py
miles/backends/megatron_utils/update_weight/__init__.py
miles/backends/megatron_utils/update_weight/common.py
miles/backends/megatron_utils/update_weight/hf_weight_iterator_base.py
miles/backends/megatron_utils/update_weight/hf_weight_iterator_bridge.py
miles/backends/megatron_utils/update_weight/hf_weight_iterator_direct.py
miles/backends/megatron_utils/update_weight/update_weight_from_tensor.py
miles/backends/megatron_utils/update_weight/update_weight_from_distributed/__init__.py
miles/backends/megatron_utils/update_weight/update_weight_from_distributed/broadcast.py
miles/backends/megatron_utils/update_weight/update_weight_from_distributed/mixin.py
miles/backends/megatron_utils/update_weight/update_weight_from_distributed/p2p.py
miles/backends/megatron_utils/update_weight/update_weight_from_distributed/p2p_transfer_utils.py
miles/backends/sglang_utils/__init__.py
miles/backends/sglang_utils/arguments.py
miles/backends/sglang_utils/sglang_config.py
miles/backends/sglang_utils/sglang_engine.py
miles/backends/training_utils/__init__.py
miles/backends/training_utils/ci_utils.py
miles/backends/training_utils/cp_utils.py
miles/backends/training_utils/data.py
miles/backends/training_utils/debug_dump.py
miles/backends/training_utils/log_utils.py
miles/backends/training_utils/loss.py
miles/backends/training_utils/parallel.py
miles/ray/__init__.py
miles/ray/actor_group.py
miles/ray/placement_group.py
miles/ray/ray_actor.py
miles/ray/rollout.py
miles/ray/train_actor.py
miles/ray/utils.py
miles/rollout/__init__.py
miles/rollout/base_types.py
miles/rollout/data_source.py
miles/rollout/sft_rollout.py
miles/rollout/sglang_rollout.py
miles/rollout/sleep_rollout.py
miles/rollout/filter_hub/__init__.py
miles/rollout/filter_hub/base_types.py
miles/rollout/filter_hub/dynamic_sampling_filters.py
miles/rollout/generate_hub/__init__.py
miles/rollout/generate_hub/agentic_tool_call.py
miles/rollout/generate_hub/benchmarkers.py
miles/rollout/generate_hub/multi_turn.py
miles/rollout/generate_hub/single_turn.py
miles/rollout/generate_utils/__init__.py
miles/rollout/generate_utils/generate_endpoint_utils.py
miles/rollout/generate_utils/openai_endpoint_utils.py
miles/rollout/generate_utils/prefill_logprobs.py
miles/rollout/generate_utils/sample_utils.py
miles/rollout/generate_utils/tool_call_utils.py
miles/rollout/inference_rollout/__init__.py
miles/rollout/inference_rollout/compatibility.py
miles/rollout/inference_rollout/inference_rollout_common.py
miles/rollout/inference_rollout/inference_rollout_eval.py
miles/rollout/inference_rollout/inference_rollout_train.py
miles/rollout/rm_hub/__init__.py
miles/rollout/rm_hub/deepscaler.py
miles/rollout/rm_hub/f1.py
miles/rollout/rm_hub/gpqa.py
miles/rollout/rm_hub/ifbench.py
miles/rollout/rm_hub/math_dapo_utils.py
miles/rollout/rm_hub/math_utils.py
miles/rollout/session/__init__.py
miles/rollout/session/linear_trajectory.py
miles/rollout/session/session_errors.py
miles/rollout/session/session_server.py
miles/rollout/session/session_types.py
miles/rollout/session/sessions.py
miles/router/__init__.py
miles/router/router.py
miles/router/middleware_hub/__init__.py
miles/router/middleware_hub/radix_tree.py
miles/router/middleware_hub/radix_tree_middleware.py
miles/true_on_policy/__init__.py
miles/true_on_policy/config.py
miles/true_on_policy/contracts.py
miles/true_on_policy/model_profiles.py
miles/true_on_policy/schema.py
miles/utils/__init__.py
miles/utils/argparse_utils.py
miles/utils/arguments.py
miles/utils/async_utils.py
miles/utils/context_utils.py
miles/utils/data.py
miles/utils/distributed_utils.py
miles/utils/dumper_utils.py
miles/utils/env_report.py
miles/utils/environ.py
miles/utils/eval_config.py
miles/utils/flops_utils.py
miles/utils/fp8_kernel.py
miles/utils/health_monitor.py
miles/utils/http_utils.py
miles/utils/iter_utils.py
miles/utils/logging_utils.py
miles/utils/mask_utils.py
miles/utils/megatron_bridge_utils.py
miles/utils/memory_utils.py
miles/utils/metric_checker.py
miles/utils/metric_utils.py
miles/utils/misc.py
miles/utils/ppo_utils.py
miles/utils/processing_utils.py
miles/utils/profile_utils.py
miles/utils/ray_utils.py
miles/utils/reloadable_process_group.py
miles/utils/replay_base.py
miles/utils/rocm_checkpoint_writer.py
miles/utils/seqlen_balancing.py
miles/utils/tensor_backper.py
miles/utils/timer.py
miles/utils/train_dump_utils.py
miles/utils/train_metric_utils.py
miles/utils/typer_utils.py
miles/utils/types.py
miles/utils/chat_template_utils/__init__.py
miles/utils/chat_template_utils/template.py
miles/utils/chat_template_utils/tito_tokenizer.py
miles/utils/chat_template_utils/token_seq_comparator.py
miles/utils/debug_utils/__init__.py
miles/utils/debug_utils/display_debug_rollout_data.py
miles/utils/debug_utils/replay_reward_fn.py
miles/utils/debug_utils/send_to_sglang.py
miles/utils/debug_utils/run_megatron/__init__.py
miles/utils/debug_utils/run_megatron/__main__.py
miles/utils/debug_utils/run_megatron/logprob_comparator.py
miles/utils/debug_utils/run_megatron/cli/__init__.py
miles/utils/debug_utils/run_megatron/cli/parallel_utils.py
miles/utils/debug_utils/run_megatron/cli/path_utils.py
miles/utils/debug_utils/run_megatron/cli/prompt_utils.py
miles/utils/debug_utils/run_megatron/cli/worker_executor.py
miles/utils/debug_utils/run_megatron/cli/commands/__init__.py
miles/utils/debug_utils/run_megatron/cli/commands/args.py
miles/utils/debug_utils/run_megatron/cli/commands/compare.py
miles/utils/debug_utils/run_megatron/cli/commands/run.py
miles/utils/debug_utils/run_megatron/cli/commands/run_and_compare.py
miles/utils/debug_utils/run_megatron/worker/__init__.py
miles/utils/debug_utils/run_megatron/worker/batch.py
miles/utils/debug_utils/run_megatron/worker/main.py
miles/utils/debug_utils/run_megatron/worker/output.py
miles/utils/debug_utils/run_megatron/worker/replay.py
miles/utils/debug_utils/run_megatron/worker/script_args.py
miles/utils/debug_utils/run_megatron/worker/top_k_print.py
miles/utils/external_utils/__init__.py
miles/utils/external_utils/command_utils.py
miles/utils/test_utils/__init__.py
miles/utils/test_utils/chat_template_verify.py
miles/utils/test_utils/mock_sglang_server.py
miles/utils/test_utils/mock_tools.py
miles/utils/test_utils/mock_trajectories.py
miles/utils/test_utils/session_verify_agent.py
miles/utils/test_utils/session_verify_runner.py
miles/utils/test_utils/uvicorn_thread_server.py
miles/utils/tracking_utils/__init__.py
miles/utils/tracking_utils/base.py
miles/utils/tracking_utils/mlflow_utils.py
miles/utils/tracking_utils/prometheus_utils.py
miles/utils/tracking_utils/tensorboard_utils.py
miles/utils/tracking_utils/wandb_utils.py
miles_megatron_plugins/__init__.py
miles_megatron_plugins/true_on_policy/__init__.py
miles_megatron_plugins/true_on_policy/attention_fa3.py
miles_megatron_plugins/true_on_policy/bias_dropout.py
miles_megatron_plugins/true_on_policy/contracts.py
miles_megatron_plugins/true_on_policy/cp_layout.py
miles_megatron_plugins/true_on_policy/debug.py
miles_megatron_plugins/true_on_policy/linear.py
miles_megatron_plugins/true_on_policy/matmul.py
miles_megatron_plugins/true_on_policy/norm.py
miles_megatron_plugins/true_on_policy/provider.py
miles_megatron_plugins/true_on_policy/rope.py
miles_megatron_plugins/true_on_policy/runtime.py
miles_megatron_plugins/true_on_policy/schema.py
miles_megatron_plugins/true_on_policy/sglang_backend.py
miles_plugins/__init__.py
miles_plugins/mbridge/__init__.py
miles_plugins/mbridge/deepseek_v32.py
miles_plugins/mbridge/glm4.py
miles_plugins/mbridge/glm4moe.py
miles_plugins/mbridge/glm4moe_lite.py
miles_plugins/mbridge/mimo.py
miles_plugins/mbridge/qwen3_5.py
miles_plugins/mbridge/qwen3_next.py
miles_plugins/megatron_bridge/__init__.py
miles_plugins/megatron_bridge/nemotron_h.py
miles_plugins/models/__init__.py
miles_plugins/models/cp_utils.py
miles_plugins/models/glm4.py
miles_plugins/models/hf_attention.py
miles_plugins/models/qwen3_5.py
miles_plugins/models/qwen3_next.py
miles_plugins/models/glm5/__init__.py
miles_plugins/models/glm5/glm5.py
miles_plugins/models/glm5/ops/__init__.py
miles_plugins/models/glm5/ops/indexer.py
miles_plugins/models/glm5/ops/sparse_mla.py
miles_plugins/models/glm5/ops/tilelang_indexer_bwd.py
miles_plugins/models/glm5/ops/tilelang_indexer_fwd.py
miles_plugins/models/glm5/ops/tilelang_sparse_mla_bwd.py
miles_plugins/models/glm5/ops/tilelang_sparse_mla_fwd.py
miles_rl.egg-info/PKG-INFO
miles_rl.egg-info/SOURCES.txt
miles_rl.egg-info/dependency_links.txt
miles_rl.egg-info/requires.txt
miles_rl.egg-info/top_level.txt
tests/test_attention_output_gate_tp.py
tests/test_chunked_gae.py
tests/test_external_rollout.py
tests/test_fsdp_import.py
tests/test_fused_experts_backward.py
third_party/Megatron-LM/megatron/core/MSC_Integration.md
third_party/Megatron-LM/megatron/core/QuickStart.md
third_party/Megatron-LM/megatron/core/README.md
third_party/Megatron-LM/megatron/core/README_STRAGGLER.md
third_party/Megatron-LM/megatron/core/__init__.py
third_party/Megatron-LM/megatron/core/_rank_utils.py
third_party/Megatron-LM/megatron/core/activations.py
third_party/Megatron-LM/megatron/core/config.py
third_party/Megatron-LM/megatron/core/config_logger.py
third_party/Megatron-LM/megatron/core/energy_monitor.py
third_party/Megatron-LM/megatron/core/enums.py
third_party/Megatron-LM/megatron/core/fp4_utils.py
third_party/Megatron-LM/megatron/core/fp8_utils.py
third_party/Megatron-LM/megatron/core/full_cuda_graph.py
third_party/Megatron-LM/megatron/core/hyper_comm_grid.py
third_party/Megatron-LM/megatron/core/inference_params.py
third_party/Megatron-LM/megatron/core/jit.py
third_party/Megatron-LM/megatron/core/model_parallel_config.py
third_party/Megatron-LM/megatron/core/msc_utils.py
third_party/Megatron-LM/megatron/core/nccl_allocator.py
third_party/Megatron-LM/megatron/core/num_microbatches_calculator.py
third_party/Megatron-LM/megatron/core/optimizer_param_scheduler.py
third_party/Megatron-LM/megatron/core/package_info.py
third_party/Megatron-LM/megatron/core/packed_seq_params.py
third_party/Megatron-LM/megatron/core/parallel_state.py
third_party/Megatron-LM/megatron/core/process_groups_config.py
third_party/Megatron-LM/megatron/core/requirements.txt
third_party/Megatron-LM/megatron/core/rerun_state_machine.py
third_party/Megatron-LM/megatron/core/safe_globals.py
third_party/Megatron-LM/megatron/core/timers.py
third_party/Megatron-LM/megatron/core/typed_torch.py
third_party/Megatron-LM/megatron/core/utils.py
third_party/Megatron-LM/megatron/core/datasets/__init__.py
third_party/Megatron-LM/megatron/core/datasets/bert_dataset.py
third_party/Megatron-LM/megatron/core/datasets/blended_dataset.py
third_party/Megatron-LM/megatron/core/datasets/blended_megatron_dataset_builder.py
third_party/Megatron-LM/megatron/core/datasets/blended_megatron_dataset_config.py
third_party/Megatron-LM/megatron/core/datasets/data_schedule.py
third_party/Megatron-LM/megatron/core/datasets/gpt_dataset.py
third_party/Megatron-LM/megatron/core/datasets/helpers.cpp
third_party/Megatron-LM/megatron/core/datasets/helpers.py
third_party/Megatron-LM/megatron/core/datasets/indexed_dataset.py
third_party/Megatron-LM/megatron/core/datasets/masked_dataset.py
third_party/Megatron-LM/megatron/core/datasets/megatron_dataset.py
third_party/Megatron-LM/megatron/core/datasets/megatron_tokenizer.py
third_party/Megatron-LM/megatron/core/datasets/multimodal_dataset.py
third_party/Megatron-LM/megatron/core/datasets/object_storage_utils.py
third_party/Megatron-LM/megatron/core/datasets/readme.md
third_party/Megatron-LM/megatron/core/datasets/t5_dataset.py
third_party/Megatron-LM/megatron/core/datasets/utils.py
third_party/Megatron-LM/megatron/core/datasets/utils_s3.py
third_party/Megatron-LM/megatron/core/dist_checkpointing/__init__.py
third_party/Megatron-LM/megatron/core/dist_checkpointing/core.py
third_party/Megatron-LM/megatron/core/dist_checkpointing/dict_utils.py
third_party/Megatron-LM/megatron/core/dist_checkpointing/exchange_utils.py
third_party/Megatron-LM/megatron/core/dist_checkpointing/mapping.py
third_party/Megatron-LM/megatron/core/dist_checkpointing/optimizer.py
third_party/Megatron-LM/megatron/core/dist_checkpointing/serialization.py
third_party/Megatron-LM/megatron/core/dist_checkpointing/state_dict_utils.py
third_party/Megatron-LM/megatron/core/dist_checkpointing/tensor_aware_state_dict.py
third_party/Megatron-LM/megatron/core/dist_checkpointing/utils.py
third_party/Megatron-LM/megatron/core/dist_checkpointing/validation.py
third_party/Megatron-LM/megatron/core/dist_checkpointing/strategies/__init__.py
third_party/Megatron-LM/megatron/core/dist_checkpointing/strategies/async_utils.py
third_party/Megatron-LM/megatron/core/dist_checkpointing/strategies/base.py
third_party/Megatron-LM/megatron/core/dist_checkpointing/strategies/cached_metadata_filesystem_reader.py
third_party/Megatron-LM/megatron/core/dist_checkpointing/strategies/checkpointable.py
third_party/Megatron-LM/megatron/core/dist_checkpointing/strategies/common.py
third_party/Megatron-LM/megatron/core/dist_checkpointing/strategies/filesystem_async.py
third_party/Megatron-LM/megatron/core/dist_checkpointing/strategies/fully_parallel.py
third_party/Megatron-LM/megatron/core/dist_checkpointing/strategies/state_dict_saver.py
third_party/Megatron-LM/megatron/core/dist_checkpointing/strategies/torch.py
third_party/Megatron-LM/megatron/core/distributed/README.md
third_party/Megatron-LM/megatron/core/distributed/__init__.py
third_party/Megatron-LM/megatron/core/distributed/data_parallel_base.py
third_party/Megatron-LM/megatron/core/distributed/distributed_data_parallel.py
third_party/Megatron-LM/megatron/core/distributed/distributed_data_parallel_config.py
third_party/Megatron-LM/megatron/core/distributed/finalize_model_grads.py
third_party/Megatron-LM/megatron/core/distributed/param_and_grad_buffer.py
third_party/Megatron-LM/megatron/core/distributed/reduce_scatter_with_fp32_accumulation.py
third_party/Megatron-LM/megatron/core/distributed/torch_fully_sharded_data_parallel.py
third_party/Megatron-LM/megatron/core/distributed/torch_fully_sharded_data_parallel_config.py
third_party/Megatron-LM/megatron/core/distributed/fsdp/__init__.py
third_party/Megatron-LM/megatron/core/distributed/fsdp/mcore_fsdp_adapter.py
third_party/Megatron-LM/megatron/core/distributed/fsdp/src/README.md
third_party/Megatron-LM/megatron/core/distributed/fsdp/src/__init__.py
third_party/Megatron-LM/megatron/core/distributed/fsdp/src/megatron_fsdp/__init__.py
third_party/Megatron-LM/megatron/core/distributed/fsdp/src/megatron_fsdp/distributed_data_parallel_config.py
third_party/Megatron-LM/megatron/core/distributed/fsdp/src/megatron_fsdp/fully_shard.py
third_party/Megatron-LM/megatron/core/distributed/fsdp/src/megatron_fsdp/megatron_fsdp.py
third_party/Megatron-LM/megatron/core/distributed/fsdp/src/megatron_fsdp/mixed_precision.py
third_party/Megatron-LM/megatron/core/distributed/fsdp/src/megatron_fsdp/package_info.py
third_party/Megatron-LM/megatron/core/distributed/fsdp/src/megatron_fsdp/param_and_grad_buffer.py
third_party/Megatron-LM/megatron/core/distributed/fsdp/src/megatron_fsdp/uneven_dtensor.py
third_party/Megatron-LM/megatron/core/distributed/fsdp/src/megatron_fsdp/utils.py
third_party/Megatron-LM/megatron/core/export/__init__.py
third_party/Megatron-LM/megatron/core/export/data_type.py
third_party/Megatron-LM/megatron/core/export/export_config.py
third_party/Megatron-LM/megatron/core/export/model_type.py
third_party/Megatron-LM/megatron/core/export/trtllm/__init__.py
third_party/Megatron-LM/megatron/core/export/trtllm/trt_model_config.py
third_party/Megatron-LM/megatron/core/export/trtllm/trt_model_type.py
third_party/Megatron-LM/megatron/core/export/trtllm/trtllm_helper.py
third_party/Megatron-LM/megatron/core/export/trtllm/trtllm_layers.py
third_party/Megatron-LM/megatron/core/export/trtllm/engine_builder/__init__.py
third_party/Megatron-LM/megatron/core/export/trtllm/engine_builder/trtllm_engine_builder.py
third_party/Megatron-LM/megatron/core/export/trtllm/model_to_trllm_mapping/__init__.py
third_party/Megatron-LM/megatron/core/export/trtllm/model_to_trllm_mapping/default_conversion_dict.py
third_party/Megatron-LM/megatron/core/export/trtllm/trtllm_weights_converter/__init__.py
third_party/Megatron-LM/megatron/core/export/trtllm/trtllm_weights_converter/distributed_trtllm_model_weights_converter.py
third_party/Megatron-LM/megatron/core/export/trtllm/trtllm_weights_converter/single_device_trtllm_model_weights_converter.py
third_party/Megatron-LM/megatron/core/export/trtllm/trtllm_weights_converter/utils.py
third_party/Megatron-LM/megatron/core/extensions/TransformerEngineMixedPrecision.md
third_party/Megatron-LM/megatron/core/extensions/__init__.py
third_party/Megatron-LM/megatron/core/extensions/kitchen.py
third_party/Megatron-LM/megatron/core/extensions/sglang.py
third_party/Megatron-LM/megatron/core/extensions/transformer_engine.py
third_party/Megatron-LM/megatron/core/extensions/transformer_engine_spec_provider.py
third_party/Megatron-LM/megatron/core/fusions/__init__.py
third_party/Megatron-LM/megatron/core/fusions/fused_bias_dropout.py
third_party/Megatron-LM/megatron/core/fusions/fused_bias_geglu.py
third_party/Megatron-LM/megatron/core/fusions/fused_bias_gelu.py
third_party/Megatron-LM/megatron/core/fusions/fused_bias_swiglu.py
third_party/Megatron-LM/megatron/core/fusions/fused_cross_entropy.py
third_party/Megatron-LM/megatron/core/fusions/fused_indices_converter.py
third_party/Megatron-LM/megatron/core/fusions/fused_layer_norm.py
third_party/Megatron-LM/megatron/core/fusions/fused_linear_cross_entropy.py
third_party/Megatron-LM/megatron/core/fusions/fused_mla_yarn_rope_apply.py
third_party/Megatron-LM/megatron/core/fusions/fused_pad_routing_map.py
third_party/Megatron-LM/megatron/core/fusions/fused_softmax.py
third_party/Megatron-LM/megatron/core/fusions/fused_weighted_squared_relu.py
third_party/Megatron-LM/megatron/core/fusions/linear_cross_entropy/__init__.py
third_party/Megatron-LM/megatron/core/fusions/linear_cross_entropy/utils.py
third_party/Megatron-LM/megatron/core/fusions/linear_cross_entropy/blackwell/__init__.py
third_party/Megatron-LM/megatron/core/fusions/linear_cross_entropy/blackwell/bwd_partial_dlogits.py
third_party/Megatron-LM/megatron/core/fusions/linear_cross_entropy/blackwell/entry.py
third_party/Megatron-LM/megatron/core/fusions/linear_cross_entropy/blackwell/fwd_mainloop.py
third_party/Megatron-LM/megatron/core/fusions/linear_cross_entropy/blackwell/triton.py
third_party/Megatron-LM/megatron/core/inference/__init__.py
third_party/Megatron-LM/megatron/core/inference/async_stream.py
third_party/Megatron-LM/megatron/core/inference/batch_dimensions_utils.py
third_party/Megatron-LM/megatron/core/inference/common_inference_params.py
third_party/Megatron-LM/megatron/core/inference/communication_utils.py
third_party/Megatron-LM/megatron/core/inference/data_parallel_inference_coordinator.py
third_party/Megatron-LM/megatron/core/inference/headers.py
third_party/Megatron-LM/megatron/core/inference/inference_client.py
third_party/Megatron-LM/megatron/core/inference/inference_request.py
third_party/Megatron-LM/megatron/core/inference/sampling_params.py
third_party/Megatron-LM/megatron/core/inference/scheduler.py
third_party/Megatron-LM/megatron/core/inference/unified_memory.py
third_party/Megatron-LM/megatron/core/inference/utils.py
third_party/Megatron-LM/megatron/core/inference/communication/torch_symm_triton/__init__.py
third_party/Megatron-LM/megatron/core/inference/communication/torch_symm_triton/barrier.py
third_party/Megatron-LM/megatron/core/inference/communication/torch_symm_triton/collectives.py
third_party/Megatron-LM/megatron/core/inference/communication/torch_symm_triton/fused_collectives.py
third_party/Megatron-LM/megatron/core/inference/communication/torch_symm_triton/multimem_asm.py
third_party/Megatron-LM/megatron/core/inference/communication/torch_symm_triton/utils.py
third_party/Megatron-LM/megatron/core/inference/contexts/__init__.py
third_party/Megatron-LM/megatron/core/inference/contexts/base_context.py
third_party/Megatron-LM/megatron/core/inference/contexts/dynamic_block_allocator.py
third_party/Megatron-LM/megatron/core/inference/contexts/dynamic_context.py
third_party/Megatron-LM/megatron/core/inference/contexts/fused_kv_append_kernel.py
third_party/Megatron-LM/megatron/core/inference/contexts/static_context.py
third_party/Megatron-LM/megatron/core/inference/contexts/attention_context/mamba_metadata.py
third_party/Megatron-LM/megatron/core/inference/contexts/attention_context/metadata_base.py
third_party/Megatron-LM/megatron/core/inference/contexts/attention_context/mha_metadata.py
third_party/Megatron-LM/megatron/core/inference/contexts/attention_context/triton/tensor_ops.py
third_party/Megatron-LM/megatron/core/inference/engines/__init__.py
third_party/Megatron-LM/megatron/core/inference/engines/abstract_engine.py
third_party/Megatron-LM/megatron/core/inference/engines/async_zmq_communicator.py
third_party/Megatron-LM/megatron/core/inference/engines/dynamic_engine.py
third_party/Megatron-LM/megatron/core/inference/engines/mcore_engine.py
third_party/Megatron-LM/megatron/core/inference/engines/static_engine.py
third_party/Megatron-LM/megatron/core/inference/model_inference_wrappers/__init__.py
third_party/Megatron-LM/megatron/core/inference/model_inference_wrappers/abstract_model_inference_wrapper.py
third_party/Megatron-LM/megatron/core/inference/model_inference_wrappers/inference_wrapper_config.py
third_party/Megatron-LM/megatron/core/inference/model_inference_wrappers/gpt/__init__.py
third_party/Megatron-LM/megatron/core/inference/model_inference_wrappers/gpt/gpt_inference_wrapper.py
third_party/Megatron-LM/megatron/core/inference/model_inference_wrappers/multimodal/vlm_inference_wrapper.py
third_party/Megatron-LM/megatron/core/inference/model_inference_wrappers/t5/__init__.py
third_party/Megatron-LM/megatron/core/inference/model_inference_wrappers/t5/t5_inference_wrapper.py
third_party/Megatron-LM/megatron/core/inference/text_generation_controllers/__init__.py
third_party/Megatron-LM/megatron/core/inference/text_generation_controllers/encoder_decoder_text_generation_controller.py
third_party/Megatron-LM/megatron/core/inference/text_generation_controllers/simple_text_generation_controller.py
third_party/Megatron-LM/megatron/core/inference/text_generation_controllers/text_generation_controller.py
third_party/Megatron-LM/megatron/core/inference/text_generation_controllers/vlm_text_generation_controller.py
third_party/Megatron-LM/megatron/core/inference/text_generation_server/__init__.py
third_party/Megatron-LM/megatron/core/inference/text_generation_server/run_mcore_engine.py
third_party/Megatron-LM/megatron/core/inference/text_generation_server/text_generation_server.py
third_party/Megatron-LM/megatron/core/inference/text_generation_server/tokenization.py
third_party/Megatron-LM/megatron/core/inference/text_generation_server/dynamic_text_gen_server/__init__.py
third_party/Megatron-LM/megatron/core/inference/text_generation_server/dynamic_text_gen_server/flask_server.py
third_party/Megatron-LM/megatron/core/inference/text_generation_server/dynamic_text_gen_server/tokenization.py
third_party/Megatron-LM/megatron/core/inference/text_generation_server/dynamic_text_gen_server/endpoints/__init__.py
third_party/Megatron-LM/megatron/core/inference/text_generation_server/dynamic_text_gen_server/endpoints/chat_completions.py
third_party/Megatron-LM/megatron/core/inference/text_generation_server/dynamic_text_gen_server/endpoints/common.py
third_party/Megatron-LM/megatron/core/inference/text_generation_server/dynamic_text_gen_server/endpoints/completions.py
third_party/Megatron-LM/megatron/core/inference/text_generation_server/dynamic_text_gen_server/endpoints/health.py
third_party/Megatron-LM/megatron/core/inference/text_generation_server/endpoints/common.py
third_party/Megatron-LM/megatron/core/inference/text_generation_server/endpoints/completions.py
third_party/Megatron-LM/megatron/core/models/__init__.py
third_party/Megatron-LM/megatron/core/models/backends.py
third_party/Megatron-LM/megatron/core/models/T5/__init__.py
third_party/Megatron-LM/megatron/core/models/T5/t5_model.py
third_party/Megatron-LM/megatron/core/models/T5/t5_spec.py
third_party/Megatron-LM/megatron/core/models/bert/__init__.py
third_party/Megatron-LM/megatron/core/models/bert/bert_layer_specs.py
third_party/Megatron-LM/megatron/core/models/bert/bert_lm_head.py
third_party/Megatron-LM/megatron/core/models/bert/bert_model.py
third_party/Megatron-LM/megatron/core/models/bert/pooler.py
third_party/Megatron-LM/megatron/core/models/common/__init__.py
third_party/Megatron-LM/megatron/core/models/common/model_chunk_schedule_plan.py
third_party/Megatron-LM/megatron/core/models/common/embeddings/__init__.py
third_party/Megatron-LM/megatron/core/models/common/embeddings/language_model_embedding.py
third_party/Megatron-LM/megatron/core/models/common/embeddings/relative_pos_embedding.py
third_party/Megatron-LM/megatron/core/models/common/embeddings/rope_utils.py
third_party/Megatron-LM/megatron/core/models/common/embeddings/rotary_pos_embedding.py
third_party/Megatron-LM/megatron/core/models/common/embeddings/yarn_rotary_pos_embedding.py
third_party/Megatron-LM/megatron/core/models/common/language_module/__init__.py
third_party/Megatron-LM/megatron/core/models/common/language_module/language_module.py
third_party/Megatron-LM/megatron/core/models/common/vision_module/__init__.py
third_party/Megatron-LM/megatron/core/models/common/vision_module/vision_module.py
third_party/Megatron-LM/megatron/core/models/gpt/__init__.py
third_party/Megatron-LM/megatron/core/models/gpt/experimental_attention_variant_module_specs.py
third_party/Megatron-LM/megatron/core/models/gpt/fine_grained_callables.py
third_party/Megatron-LM/megatron/core/models/gpt/gpt_layer_specs.py
third_party/Megatron-LM/megatron/core/models/gpt/gpt_model.py
third_party/Megatron-LM/megatron/core/models/gpt/moe_module_specs.py
third_party/Megatron-LM/megatron/core/models/gpt/heterogeneous/heterogeneous_layer_specs.py
third_party/Megatron-LM/megatron/core/models/huggingface/__init__.py
third_party/Megatron-LM/megatron/core/models/huggingface/clip_model.py
third_party/Megatron-LM/megatron/core/models/huggingface/module.py
third_party/Megatron-LM/megatron/core/models/huggingface/qwen_model.py
third_party/Megatron-LM/megatron/core/models/mamba/__init__.py
third_party/Megatron-LM/megatron/core/models/mamba/mamba_layer_specs.py
third_party/Megatron-LM/megatron/core/models/mamba/mamba_model.py
third_party/Megatron-LM/megatron/core/models/mimo/README.md
third_party/Megatron-LM/megatron/core/models/mimo/__init__.py
third_party/Megatron-LM/megatron/core/models/mimo/config/__init__.py
third_party/Megatron-LM/megatron/core/models/mimo/config/base_configs.py
third_party/Megatron-LM/megatron/core/models/mimo/model/__init__.py
third_party/Megatron-LM/megatron/core/models/mimo/model/base.py
third_party/Megatron-LM/megatron/core/models/mimo/submodules/audio.py
third_party/Megatron-LM/megatron/core/models/mimo/submodules/base.py
third_party/Megatron-LM/megatron/core/models/mimo/submodules/vision.py
third_party/Megatron-LM/megatron/core/models/multimodal/__init__.py
third_party/Megatron-LM/megatron/core/models/multimodal/context_parallel.py
third_party/Megatron-LM/megatron/core/models/multimodal/llava_model.py
third_party/Megatron-LM/megatron/core/models/multimodal/llava_spec.py
third_party/Megatron-LM/megatron/core/models/vision/__init__.py
third_party/Megatron-LM/megatron/core/models/vision/clip_vit_model.py
third_party/Megatron-LM/megatron/core/models/vision/multimodal_projector.py
third_party/Megatron-LM/megatron/core/models/vision/radio.py
third_party/Megatron-LM/megatron/core/models/vision/vit_layer_specs.py
third_party/Megatron-LM/megatron/core/optimizer/__init__.py
third_party/Megatron-LM/megatron/core/optimizer/clip_grads.py
third_party/Megatron-LM/megatron/core/optimizer/distrib_optimizer.py
third_party/Megatron-LM/megatron/core/optimizer/fused_adam_patch.py
third_party/Megatron-LM/megatron/core/optimizer/grad_scaler.py
third_party/Megatron-LM/megatron/core/optimizer/layer_wise_optimizer.py
third_party/Megatron-LM/megatron/core/optimizer/muon.py
third_party/Megatron-LM/megatron/core/optimizer/optimizer.py
third_party/Megatron-LM/megatron/core/optimizer/optimizer_config.py
third_party/Megatron-LM/megatron/core/optimizer/qk_clip.py
third_party/Megatron-LM/megatron/core/optimizer/cpu_offloading/README.md
third_party/Megatron-LM/megatron/core/optimizer/cpu_offloading/__init__.py
third_party/Megatron-LM/megatron/core/optimizer/cpu_offloading/hybrid_optimizer.py
third_party/Megatron-LM/megatron/core/optimizer/cpu_offloading/optimizer_state_offloader.py
third_party/Megatron-LM/megatron/core/pipeline_parallel/__init__.py
third_party/Megatron-LM/megatron/core/pipeline_parallel/bridge_communicator.py
third_party/Megatron-LM/megatron/core/pipeline_parallel/combined_1f1b.py
third_party/Megatron-LM/megatron/core/pipeline_parallel/fine_grained_activation_offload.py
third_party/Megatron-LM/megatron/core/pipeline_parallel/hybrid_cp_schedule.py
third_party/Megatron-LM/megatron/core/pipeline_parallel/multimodule_communicator.py
third_party/Megatron-LM/megatron/core/pipeline_parallel/p2p_communication.py
third_party/Megatron-LM/megatron/core/pipeline_parallel/schedules.py
third_party/Megatron-LM/megatron/core/pipeline_parallel/utils.py
third_party/Megatron-LM/megatron/core/post_training/__init__.py
third_party/Megatron-LM/megatron/core/post_training/modelopt/__init__.py
third_party/Megatron-LM/megatron/core/post_training/modelopt/layers.py
third_party/Megatron-LM/megatron/core/post_training/modelopt/gpt/__init__.py
third_party/Megatron-LM/megatron/core/post_training/modelopt/gpt/model_specs.py
third_party/Megatron-LM/megatron/core/post_training/modelopt/gpt/state_dict_hooks.py
third_party/Megatron-LM/megatron/core/post_training/modelopt/mamba/__init__.py
third_party/Megatron-LM/megatron/core/post_training/modelopt/mamba/model_specs.py
third_party/Megatron-LM/megatron/core/quantization/__init__.py
third_party/Megatron-LM/megatron/core/quantization/quant_config.py
third_party/Megatron-LM/megatron/core/quantization/utils.py
third_party/Megatron-LM/megatron/core/resharding/__init__.py
third_party/Megatron-LM/megatron/core/resharding/execution.py
third_party/Megatron-LM/megatron/core/resharding/planner.py
third_party/Megatron-LM/megatron/core/resharding/refit.py
third_party/Megatron-LM/megatron/core/resharding/utils.py
third_party/Megatron-LM/megatron/core/resharding/copy_services/__init__.py
third_party/Megatron-LM/megatron/core/resharding/copy_services/base.py
third_party/Megatron-LM/megatron/core/resharding/copy_services/gloo_copy_service.py
third_party/Megatron-LM/megatron/core/resharding/copy_services/nccl_copy_service.py
third_party/Megatron-LM/megatron/core/ssm/__init__.py
third_party/Megatron-LM/megatron/core/ssm/gated_delta_net.py
third_party/Megatron-LM/megatron/core/ssm/mamba_block.py
third_party/Megatron-LM/megatron/core/ssm/mamba_context_parallel.py
third_party/Megatron-LM/megatron/core/ssm/mamba_hybrid_layer_allocation.py
third_party/Megatron-LM/megatron/core/ssm/mamba_layer.py
third_party/Megatron-LM/megatron/core/ssm/mamba_mixer.py
third_party/Megatron-LM/megatron/core/ssm/mlp_layer.py
third_party/Megatron-LM/megatron/core/ssm/triton_cache_manager.py
third_party/Megatron-LM/megatron/core/tensor_parallel/__init__.py
third_party/Megatron-LM/megatron/core/tensor_parallel/cross_entropy.py
third_party/Megatron-LM/megatron/core/tensor_parallel/data.py
third_party/Megatron-LM/megatron/core/tensor_parallel/inference_layers.py
third_party/Megatron-LM/megatron/core/tensor_parallel/layers.py
third_party/Megatron-LM/megatron/core/tensor_parallel/mappings.py
third_party/Megatron-LM/megatron/core/tensor_parallel/matmul_tp_inv.py
third_party/Megatron-LM/megatron/core/tensor_parallel/random.py
third_party/Megatron-LM/megatron/core/tensor_parallel/utils.py
third_party/Megatron-LM/megatron/core/tokenizers/__init__.py
third_party/Megatron-LM/megatron/core/tokenizers/base_tokenizer.py
third_party/Megatron-LM/megatron/core/tokenizers/megatron_tokenizer.py
third_party/Megatron-LM/megatron/core/tokenizers/text/__init__.py
third_party/Megatron-LM/megatron/core/tokenizers/text/text_tokenizer.py
third_party/Megatron-LM/megatron/core/tokenizers/text/libraries/__init__.py
third_party/Megatron-LM/megatron/core/tokenizers/text/libraries/abstract_tokenizer.py
third_party/Megatron-LM/megatron/core/tokenizers/text/libraries/bytelevel_tokenizer.py
third_party/Megatron-LM/megatron/core/tokenizers/text/libraries/chat_template.py
third_party/Megatron-LM/megatron/core/tokenizers/text/libraries/huggingface_tokenizer.py
third_party/Megatron-LM/megatron/core/tokenizers/text/libraries/megatron_hf_tokenizer.py
third_party/Megatron-LM/megatron/core/tokenizers/text/libraries/null_tokenizer.py
third_party/Megatron-LM/megatron/core/tokenizers/text/libraries/sentencepiece_tokenizer.py
third_party/Megatron-LM/megatron/core/tokenizers/text/libraries/tiktoken_tokenizer.py
third_party/Megatron-LM/megatron/core/tokenizers/text/models/__init__.py
third_party/Megatron-LM/megatron/core/tokenizers/text/models/bert_tokenizer.py
third_party/Megatron-LM/megatron/core/tokenizers/text/models/default_tokenizer.py
third_party/Megatron-LM/megatron/core/tokenizers/text/models/gpt_tokenizer.py
third_party/Megatron-LM/megatron/core/tokenizers/text/models/mamba_tokenizer.py
third_party/Megatron-LM/megatron/core/tokenizers/text/models/t5_tokenizer.py
third_party/Megatron-LM/megatron/core/tokenizers/text/utils/build_tokenizer.py
third_party/Megatron-LM/megatron/core/transformer/__init__.py
third_party/Megatron-LM/megatron/core/transformer/attention.py
third_party/Megatron-LM/megatron/core/transformer/cuda_graphs.py
third_party/Megatron-LM/megatron/core/transformer/dot_product_attention.py
third_party/Megatron-LM/megatron/core/transformer/dot_product_attention_context_parallel.py
third_party/Megatron-LM/megatron/core/transformer/enums.py
third_party/Megatron-LM/megatron/core/transformer/fsdp_dtensor_checkpoint.py
third_party/Megatron-LM/megatron/core/transformer/identity_op.py
third_party/Megatron-LM/megatron/core/transformer/linear_cross_entropy.py
third_party/Megatron-LM/megatron/core/transformer/mlp.py
third_party/Megatron-LM/megatron/core/transformer/module.py
third_party/Megatron-LM/megatron/core/transformer/multi_latent_attention.py
third_party/Megatron-LM/megatron/core/transformer/multi_token_prediction.py
third_party/Megatron-LM/megatron/core/transformer/pipeline_parallel_layer_layout.py
third_party/Megatron-LM/megatron/core/transformer/spec_utils.py
third_party/Megatron-LM/megatron/core/transformer/torch_layer_norm.py
third_party/Megatron-LM/megatron/core/transformer/torch_norm.py
third_party/Megatron-LM/megatron/core/transformer/transformer_block.py
third_party/Megatron-LM/megatron/core/transformer/transformer_config.py
third_party/Megatron-LM/megatron/core/transformer/transformer_layer.py
third_party/Megatron-LM/megatron/core/transformer/utils.py
third_party/Megatron-LM/megatron/core/transformer/custom_layers/__init__.py
third_party/Megatron-LM/megatron/core/transformer/custom_layers/batch_invariant_kernels.py
third_party/Megatron-LM/megatron/core/transformer/custom_layers/transformer_engine.py
third_party/Megatron-LM/megatron/core/transformer/experimental_attention_variant/absorbed_mla.py
third_party/Megatron-LM/megatron/core/transformer/experimental_attention_variant/dsa.py
third_party/Megatron-LM/megatron/core/transformer/heterogeneous/heterogeneous_config.py
third_party/Megatron-LM/megatron/core/transformer/heterogeneous/linear_replacements.py
third_party/Megatron-LM/megatron/core/transformer/moe/README.md
third_party/Megatron-LM/megatron/core/transformer/moe/__init__.py
third_party/Megatron-LM/megatron/core/transformer/moe/experts.py
third_party/Megatron-LM/megatron/core/transformer/moe/fused_a2a.py
third_party/Megatron-LM/megatron/core/transformer/moe/grouped_gemm_util.py
third_party/Megatron-LM/megatron/core/transformer/moe/moe_layer.py
third_party/Megatron-LM/megatron/core/transformer/moe/moe_utils.py
third_party/Megatron-LM/megatron/core/transformer/moe/router.py
third_party/Megatron-LM/megatron/core/transformer/moe/router_replay.py
third_party/Megatron-LM/megatron/core/transformer/moe/shared_experts.py
third_party/Megatron-LM/megatron/core/transformer/moe/token_dispatcher.py
third_party/Megatron-LM/megatron/core/transformer/moe/upcycling_utils.py
third_party/Megatron-LM/megatron/legacy/indexer.py
third_party/Megatron-LM/megatron/legacy/data/__init__.py
third_party/Megatron-LM/megatron/legacy/data/autoaugment.py
third_party/Megatron-LM/megatron/legacy/data/biencoder_dataset_utils.py
third_party/Megatron-LM/megatron/legacy/data/dataset_utils.py
third_party/Megatron-LM/megatron/legacy/data/ict_dataset.py
third_party/Megatron-LM/megatron/legacy/data/image_folder.py
third_party/Megatron-LM/megatron/legacy/data/multimodal_dataset.py
third_party/Megatron-LM/megatron/legacy/data/orqa_wiki_dataset.py
third_party/Megatron-LM/megatron/legacy/data/realm_dataset_utils.py
third_party/Megatron-LM/megatron/legacy/data/realm_index.py
third_party/Megatron-LM/megatron/legacy/data/vit_dataset.py
third_party/Megatron-LM/megatron/legacy/fp16_deprecated/loss_scaler.py
third_party/Megatron-LM/megatron/legacy/fused_kernels/__init__.py
third_party/Megatron-LM/megatron/legacy/fused_kernels/compat.h
third_party/Megatron-LM/megatron/legacy/fused_kernels/type_shim.h
third_party/Megatron-LM/megatron/legacy/fused_kernels/tests/__init__.py
third_party/Megatron-LM/megatron/legacy/fused_kernels/tests/test_fused_kernels.py
third_party/Megatron-LM/megatron/legacy/model/__init__.py
third_party/Megatron-LM/megatron/legacy/model/bert_model.py
third_party/Megatron-LM/megatron/legacy/model/biencoder_model.py
third_party/Megatron-LM/megatron/legacy/model/classification.py
third_party/Megatron-LM/megatron/legacy/model/enums.py
third_party/Megatron-LM/megatron/legacy/model/fused_bias_gelu.py
third_party/Megatron-LM/megatron/legacy/model/fused_layer_norm.py
third_party/Megatron-LM/megatron/legacy/model/fused_softmax.py
third_party/Megatron-LM/megatron/legacy/model/gpt_model.py
third_party/Megatron-LM/megatron/legacy/model/language_model.py
third_party/Megatron-LM/megatron/legacy/model/module.py
third_party/Megatron-LM/megatron/legacy/model/multiple_choice.py
third_party/Megatron-LM/megatron/legacy/model/realm_model.py
third_party/Megatron-LM/megatron/legacy/model/rms_norm.py
third_party/Megatron-LM/megatron/legacy/model/t5_model.py
third_party/Megatron-LM/megatron/legacy/model/transformer.py
third_party/Megatron-LM/megatron/legacy/model/utils.py
third_party/Megatron-LM/megatron/legacy/model/vision/classification.py
third_party/Megatron-LM/megatron/legacy/model/vision/dino.py
third_party/Megatron-LM/megatron/legacy/model/vision/esvit_swin_backbone.py
third_party/Megatron-LM/megatron/legacy/model/vision/inpainting.py
third_party/Megatron-LM/megatron/legacy/model/vision/knn_monitor.py
third_party/Megatron-LM/megatron/legacy/model/vision/mit_backbone.py
third_party/Megatron-LM/megatron/legacy/model/vision/swin_backbone.py
third_party/Megatron-LM/megatron/legacy/model/vision/utils.py
third_party/Megatron-LM/megatron/legacy/model/vision/vit_backbone.py
third_party/Megatron-LM/megatron/legacy/mpu/tests/__init__.py
third_party/Megatron-LM/megatron/legacy/mpu/tests/commons.py
third_party/Megatron-LM/megatron/legacy/mpu/tests/test_cross_entropy.py
third_party/Megatron-LM/megatron/legacy/mpu/tests/test_data.py
third_party/Megatron-LM/megatron/legacy/mpu/tests/test_initialize.py
third_party/Megatron-LM/megatron/legacy/mpu/tests/test_layers.py
third_party/Megatron-LM/megatron/legacy/mpu/tests/test_random.py
third_party/Megatron-LM/megatron/post_training/__init__.py
third_party/Megatron-LM/megatron/post_training/arguments.py
third_party/Megatron-LM/megatron/post_training/checkpointing.py
third_party/Megatron-LM/megatron/post_training/generate.py
third_party/Megatron-LM/megatron/post_training/loss_func.py
third_party/Megatron-LM/megatron/post_training/model_builder.py
third_party/Megatron-LM/megatron/post_training/non_loss_data_func.py
third_party/Megatron-LM/megatron/post_training/utils.py
third_party/Megatron-LM/megatron/rl/README.md
third_party/Megatron-LM/megatron/rl/__init__.py
third_party/Megatron-LM/megatron/rl/logging.py
third_party/Megatron-LM/megatron/rl/parallel_utils.py
third_party/Megatron-LM/megatron/rl/rl_utils.py
third_party/Megatron-LM/megatron/rl/sequence_packing_utils.py
third_party/Megatron-LM/megatron/rl/agent/__init__.py
third_party/Megatron-LM/megatron/rl/agent/api.py
third_party/Megatron-LM/megatron/rl/agent/huggingface_dataset_agent.py
third_party/Megatron-LM/megatron/rl/agent/pass_at_evaluation_agent.py
third_party/Megatron-LM/megatron/rl/agent/remote_agent.py
third_party/Megatron-LM/megatron/rl/agent/reward_only_agent.py
third_party/Megatron-LM/megatron/rl/agent/weighted_multi_task.py
third_party/Megatron-LM/megatron/rl/inference/__init__.py
third_party/Megatron-LM/megatron/rl/inference/api.py
third_party/Megatron-LM/megatron/rl/inference/chat_templates.py
third_party/Megatron-LM/megatron/rl/inference/inference_interface.py
third_party/Megatron-LM/megatron/rl/inference/megatron.py
third_party/Megatron-LM/megatron/rl/server/__init__.py
third_party/Megatron-LM/megatron/rl/server/api.py
third_party/Megatron-LM/megatron/rl/server/agent/__init__.py
third_party/Megatron-LM/megatron/rl/server/agent/fastapi_env_server.py
third_party/Megatron-LM/megatron/rl/server/inference/__init__.py
third_party/Megatron-LM/megatron/rl/server/inference/inference_interface_server.py
third_party/Megatron-LM/megatron/training/__init__.py
third_party/Megatron-LM/megatron/training/argument_utils.py
third_party/Megatron-LM/megatron/training/arguments.py
third_party/Megatron-LM/megatron/training/async_utils.py
third_party/Megatron-LM/megatron/training/checkpointing.py
third_party/Megatron-LM/megatron/training/common_config.py
third_party/Megatron-LM/megatron/training/dgrad_logging.py
third_party/Megatron-LM/megatron/training/dist_signal_handler.py
third_party/Megatron-LM/megatron/training/ft_integration.py
third_party/Megatron-LM/megatron/training/global_vars.py
third_party/Megatron-LM/megatron/training/initialize.py
third_party/Megatron-LM/megatron/training/inprocess_restart.py
third_party/Megatron-LM/megatron/training/log_handler.py
third_party/Megatron-LM/megatron/training/one_logger_utils.py
third_party/Megatron-LM/megatron/training/resilience_config.py
third_party/Megatron-LM/megatron/training/theoretical_memory_usage.py
third_party/Megatron-LM/megatron/training/training.py
third_party/Megatron-LM/megatron/training/training_config.py
third_party/Megatron-LM/megatron/training/utils.py
third_party/Megatron-LM/megatron/training/wandb_utils.py
third_party/Megatron-LM/megatron/training/yaml_arguments.py
third_party/Megatron-LM/megatron/training/datasets/README.md
third_party/Megatron-LM/megatron/training/datasets/__init__.py
third_party/Megatron-LM/megatron/training/datasets/data_samplers.py
third_party/Megatron-LM/megatron/training/datasets/fim_dataset.py
third_party/Megatron-LM/megatron/training/datasets/sft_dataset.py
third_party/Megatron-LM/megatron/training/tokenizer/__init__.py
third_party/Megatron-LM/megatron/training/tokenizer/bert_tokenization.py
third_party/Megatron-LM/megatron/training/tokenizer/gpt2_tokenization.py
third_party/Megatron-LM/megatron/training/tokenizer/multimodal_tokenizer.py
third_party/Megatron-LM/megatron/training/tokenizer/sft_tokenizer.py
third_party/Megatron-LM/megatron/training/tokenizer/tokenizer.py
third_party/sglang/python/sglang/README.md
third_party/sglang/python/sglang/__init__.py
third_party/sglang/python/sglang/_mps_stub.py
third_party/sglang/python/sglang/_triton_stub.py
third_party/sglang/python/sglang/bench_offline_throughput.py
third_party/sglang/python/sglang/bench_one_batch.py
third_party/sglang/python/sglang/bench_one_batch_server.py
third_party/sglang/python/sglang/bench_serving.py
third_party/sglang/python/sglang/check_env.py
third_party/sglang/python/sglang/compile_deep_gemm.py
third_party/sglang/python/sglang/global_config.py
third_party/sglang/python/sglang/kernel_api_logging.py
third_party/sglang/python/sglang/launch_server.py
third_party/sglang/python/sglang/profiler.py
third_party/sglang/python/sglang/utils.py
third_party/sglang/python/sglang/version.py
third_party/sglang/python/sglang/benchmark/__init__.py
third_party/sglang/python/sglang/benchmark/bench_utils.py
third_party/sglang/python/sglang/benchmark/utils.py
third_party/sglang/python/sglang/benchmark/datasets/__init__.py
third_party/sglang/python/sglang/benchmark/datasets/common.py
third_party/sglang/python/sglang/benchmark/datasets/custom.py
third_party/sglang/python/sglang/benchmark/datasets/generated_shared_prefix.py
third_party/sglang/python/sglang/benchmark/datasets/image.py
third_party/sglang/python/sglang/benchmark/datasets/longbench_v2.py
third_party/sglang/python/sglang/benchmark/datasets/mmmu.py
third_party/sglang/python/sglang/benchmark/datasets/mooncake.py
third_party/sglang/python/sglang/benchmark/datasets/openai_dataset.py
third_party/sglang/python/sglang/benchmark/datasets/random.py
third_party/sglang/python/sglang/benchmark/datasets/sharegpt.py
third_party/sglang/python/sglang/cli/__init__.py
third_party/sglang/python/sglang/cli/generate.py
third_party/sglang/python/sglang/cli/killall.py
third_party/sglang/python/sglang/cli/main.py
third_party/sglang/python/sglang/cli/serve.py
third_party/sglang/python/sglang/cli/utils.py
third_party/sglang/python/sglang/eval/llama3_eval.py
third_party/sglang/python/sglang/eval/loogle_eval.py
third_party/sglang/python/sglang/jit_kernel/__init__.py
third_party/sglang/python/sglang/jit_kernel/__main__.py
third_party/sglang/python/sglang/jit_kernel/add_constant.py
third_party/sglang/python/sglang/jit_kernel/all_reduce.py
third_party/sglang/python/sglang/jit_kernel/awq_dequantize.py
third_party/sglang/python/sglang/jit_kernel/awq_marlin_repack.py
third_party/sglang/python/sglang/jit_kernel/cast.py
third_party/sglang/python/sglang/jit_kernel/clamp_position.py
third_party/sglang/python/sglang/jit_kernel/concat_mla.py
third_party/sglang/python/sglang/jit_kernel/cutedsl_gdn.py
third_party/sglang/python/sglang/jit_kernel/cutedsl_kda.py
third_party/sglang/python/sglang/jit_kernel/flash_attention_v4.py
third_party/sglang/python/sglang/jit_kernel/fused_metadata_copy.py
third_party/sglang/python/sglang/jit_kernel/fused_qknorm_rope.py
third_party/sglang/python/sglang/jit_kernel/fused_store_index_cache.py
third_party/sglang/python/sglang/jit_kernel/gptq_marlin.py
third_party/sglang/python/sglang/jit_kernel/gptq_marlin_repack.py
third_party/sglang/python/sglang/jit_kernel/hadamard.py
third_party/sglang/python/sglang/jit_kernel/hicache.py
third_party/sglang/python/sglang/jit_kernel/hisparse.py
third_party/sglang/python/sglang/jit_kernel/kvcache.py
third_party/sglang/python/sglang/jit_kernel/moe_align.py
third_party/sglang/python/sglang/jit_kernel/moe_lora_align.py
third_party/sglang/python/sglang/jit_kernel/moe_wna16_marlin.py
third_party/sglang/python/sglang/jit_kernel/ngram_corpus.py
third_party/sglang/python/sglang/jit_kernel/ngram_embedding.py
third_party/sglang/python/sglang/jit_kernel/norm.py
third_party/sglang/python/sglang/jit_kernel/nvfp4.py
third_party/sglang/python/sglang/jit_kernel/per_tensor_quant_fp8.py
third_party/sglang/python/sglang/jit_kernel/per_token_group_quant_8bit.py
third_party/sglang/python/sglang/jit_kernel/resolve_future_token_ids.py
third_party/sglang/python/sglang/jit_kernel/rope.py
third_party/sglang/python/sglang/jit_kernel/timestep_embedding.py
third_party/sglang/python/sglang/jit_kernel/utils.py
third_party/sglang/python/sglang/jit_kernel/benchmark/bench_awq_dequantize.py
third_party/sglang/python/sglang/jit_kernel/benchmark/bench_cast.py
third_party/sglang/python/sglang/jit_kernel/benchmark/bench_clamp_position.py
third_party/sglang/python/sglang/jit_kernel/benchmark/bench_concat_mla.py
third_party/sglang/python/sglang/jit_kernel/benchmark/bench_custom_all_reduce.py
third_party/sglang/python/sglang/jit_kernel/benchmark/bench_fused_qknorm_rope.py
third_party/sglang/python/sglang/jit_kernel/benchmark/bench_hadamard.py
third_party/sglang/python/sglang/jit_kernel/benchmark/bench_hicache.py
third_party/sglang/python/sglang/jit_kernel/benchmark/bench_norm.py
third_party/sglang/python/sglang/jit_kernel/benchmark/bench_nvfp4_blockwise_moe.py
third_party/sglang/python/sglang/jit_kernel/benchmark/bench_nvfp4_quant.py
third_party/sglang/python/sglang/jit_kernel/benchmark/bench_nvfp4_scaled_mm.py
third_party/sglang/python/sglang/jit_kernel/benchmark/bench_per_tensor_quant_fp8.py
third_party/sglang/python/sglang/jit_kernel/benchmark/bench_per_token_group_quant_8bit.py
third_party/sglang/python/sglang/jit_kernel/benchmark/bench_qknorm.py
third_party/sglang/python/sglang/jit_kernel/benchmark/bench_qknorm_across_heads.py
third_party/sglang/python/sglang/jit_kernel/benchmark/bench_renorm.py
third_party/sglang/python/sglang/jit_kernel/benchmark/bench_resolve_future_token_ids.py
third_party/sglang/python/sglang/jit_kernel/benchmark/bench_rope.py
third_party/sglang/python/sglang/jit_kernel/benchmark/bench_store_cache.py
third_party/sglang/python/sglang/jit_kernel/benchmark/utils.py
third_party/sglang/python/sglang/jit_kernel/benchmark/diffusion/bench_diffusion_nvfp4_scaled_mm.py
third_party/sglang/python/sglang/jit_kernel/benchmark/diffusion/bench_fused_norm_scale_shift.py
third_party/sglang/python/sglang/jit_kernel/benchmark/diffusion/bench_norm_impls.py
third_party/sglang/python/sglang/jit_kernel/benchmark/diffusion/bench_qknorm_rope.py
third_party/sglang/python/sglang/jit_kernel/benchmark/diffusion/bench_qwen_image_modulation.py
third_party/sglang/python/sglang/jit_kernel/benchmark/diffusion/diffusion_nvfp4_shapes.json
third_party/sglang/python/sglang/jit_kernel/csrc/add_constant.cuh
third_party/sglang/python/sglang/jit_kernel/csrc/hicache.cuh
third_party/sglang/python/sglang/jit_kernel/csrc/hisparse.cuh
third_party/sglang/python/sglang/jit_kernel/csrc/ngram_embedding.cuh
third_party/sglang/python/sglang/jit_kernel/csrc/diffusion/qknorm_rope.cuh
third_party/sglang/python/sglang/jit_kernel/csrc/diffusion/timestep_embedding.cuh
third_party/sglang/python/sglang/jit_kernel/csrc/distributed/custom_all_reduce_base.cuh
third_party/sglang/python/sglang/jit_kernel/csrc/distributed/custom_all_reduce_pull.cuh
third_party/sglang/python/sglang/jit_kernel/csrc/distributed/custom_all_reduce_push.cuh
third_party/sglang/python/sglang/jit_kernel/csrc/elementwise/cast.cuh
third_party/sglang/python/sglang/jit_kernel/csrc/elementwise/clamp_position.cuh
third_party/sglang/python/sglang/jit_kernel/csrc/elementwise/concat_mla.cuh
third_party/sglang/python/sglang/jit_kernel/csrc/elementwise/fused_add_rmsnorm.cuh
third_party/sglang/python/sglang/jit_kernel/csrc/elementwise/fused_metadata_copy.cuh
third_party/sglang/python/sglang/jit_kernel/csrc/elementwise/fused_qknorm_rope.cuh
third_party/sglang/python/sglang/jit_kernel/csrc/elementwise/kvcache.cuh
third_party/sglang/python/sglang/jit_kernel/csrc/elementwise/pos_enc.cuh
third_party/sglang/python/sglang/jit_kernel/csrc/elementwise/qknorm.cuh
third_party/sglang/python/sglang/jit_kernel/csrc/elementwise/qknorm_across_heads.cuh
third_party/sglang/python/sglang/jit_kernel/csrc/elementwise/resolve_future_token_ids.cuh
third_party/sglang/python/sglang/jit_kernel/csrc/elementwise/rmsnorm.cuh
third_party/sglang/python/sglang/jit_kernel/csrc/elementwise/rope.cuh
third_party/sglang/python/sglang/jit_kernel/csrc/fast-hadamard-transform/code_gen.py
third_party/sglang/python/sglang/jit_kernel/csrc/fast-hadamard-transform/fast_hadamard_transform.h
third_party/sglang/python/sglang/jit_kernel/csrc/fast-hadamard-transform/fast_hadamard_transform_common.h
third_party/sglang/python/sglang/jit_kernel/csrc/fast-hadamard-transform/fast_hadamard_transform_special.h
third_party/sglang/python/sglang/jit_kernel/csrc/fast-hadamard-transform/hadamard_jit.cuh
third_party/sglang/python/sglang/jit_kernel/csrc/fast-hadamard-transform/static_switch.h
third_party/sglang/python/sglang/jit_kernel/csrc/gemm/awq_dequantize.cuh
third_party/sglang/python/sglang/jit_kernel/csrc/gemm/per_tensor_quant_fp8.cuh
third_party/sglang/python/sglang/jit_kernel/csrc/gemm/per_token_group_quant_8bit.cuh
third_party/sglang/python/sglang/jit_kernel/csrc/gemm/marlin/awq_marlin_repack.cuh
third_party/sglang/python/sglang/jit_kernel/csrc/gemm/marlin/dequant.h
third_party/sglang/python/sglang/jit_kernel/csrc/gemm/marlin/gptq_marlin.cuh
third_party/sglang/python/sglang/jit_kernel/csrc/gemm/marlin/gptq_marlin_repack.cuh
third_party/sglang/python/sglang/jit_kernel/csrc/gemm/marlin/kernel.h
third_party/sglang/python/sglang/jit_kernel/csrc/gemm/marlin/marlin.cuh
third_party/sglang/python/sglang/jit_kernel/csrc/gemm/marlin/marlin_dtypes.cuh
third_party/sglang/python/sglang/jit_kernel/csrc/gemm/marlin/marlin_template.h
third_party/sglang/python/sglang/jit_kernel/csrc/gemm/marlin_moe/kernel.h
third_party/sglang/python/sglang/jit_kernel/csrc/gemm/marlin_moe/marlin_template.h
third_party/sglang/python/sglang/jit_kernel/csrc/gemm/marlin_moe/moe_wna16_marlin.cuh
third_party/sglang/python/sglang/jit_kernel/csrc/gemm/nvfp4/nvfp4_expert_quant.cuh
third_party/sglang/python/sglang/jit_kernel/csrc/gemm/nvfp4/nvfp4_quant.cuh
third_party/sglang/python/sglang/jit_kernel/csrc/gemm/nvfp4/nvfp4_quant_entry.cuh
third_party/sglang/python/sglang/jit_kernel/csrc/gemm/nvfp4/nvfp4_quant_kernels.cuh
third_party/sglang/python/sglang/jit_kernel/csrc/gemm/nvfp4/nvfp4_scaled_mm_common.cuh
third_party/sglang/python/sglang/jit_kernel/csrc/gemm/nvfp4/nvfp4_scaled_mm_entry.cuh
third_party/sglang/python/sglang/jit_kernel/csrc/gemm/nvfp4/nvfp4_scaled_mm_kernels.cuh
third_party/sglang/python/sglang/jit_kernel/csrc/gemm/nvfp4/nvfp4_scaled_mm_sm100.cuh
third_party/sglang/python/sglang/jit_kernel/csrc/gemm/nvfp4/nvfp4_scaled_mm_sm120.cuh
third_party/sglang/python/sglang/jit_kernel/csrc/lora/moe_lora_align_kernel.cu
third_party/sglang/python/sglang/jit_kernel/csrc/moe/moe_align_kernel.cu
third_party/sglang/python/sglang/jit_kernel/csrc/moe/nvfp4_blockwise_moe.cuh
third_party/sglang/python/sglang/jit_kernel/csrc/ngram_corpus/ngram.cpp
third_party/sglang/python/sglang/jit_kernel/csrc/ngram_corpus/ngram.h
third_party/sglang/python/sglang/jit_kernel/csrc/ngram_corpus/ngram_corpus_ffi.cpp
third_party/sglang/python/sglang/jit_kernel/csrc/ngram_corpus/param.h
third_party/sglang/python/sglang/jit_kernel/csrc/ngram_corpus/queue.h
third_party/sglang/python/sglang/jit_kernel/csrc/ngram_corpus/result.cpp
third_party/sglang/python/sglang/jit_kernel/csrc/ngram_corpus/result.h
third_party/sglang/python/sglang/jit_kernel/csrc/ngram_corpus/trie.cpp
third_party/sglang/python/sglang/jit_kernel/csrc/ngram_corpus/trie.h
third_party/sglang/python/sglang/jit_kernel/csrc/nsa/fused_store_index_cache.cuh
third_party/sglang/python/sglang/jit_kernel/diffusion/qknorm_rope.py
third_party/sglang/python/sglang/jit_kernel/diffusion/cutedsl/norm_tanh_mul_add_norm_scale.py
third_party/sglang/python/sglang/jit_kernel/diffusion/cutedsl/scale_residual_norm_scale_shift.py
third_party/sglang/python/sglang/jit_kernel/diffusion/cutedsl/utils.py
third_party/sglang/python/sglang/jit_kernel/diffusion/cutedsl/common/norm_fusion.py
third_party/sglang/python/sglang/jit_kernel/diffusion/cutedsl/common/reduce.py
third_party/sglang/python/sglang/jit_kernel/diffusion/triton/mps_fallback.py
third_party/sglang/python/sglang/jit_kernel/diffusion/triton/norm.py
third_party/sglang/python/sglang/jit_kernel/diffusion/triton/npu_fallback.py
third_party/sglang/python/sglang/jit_kernel/diffusion/triton/rmsnorm_onepass.py
third_party/sglang/python/sglang/jit_kernel/diffusion/triton/rotary.py
third_party/sglang/python/sglang/jit_kernel/diffusion/triton/scale_shift.py
third_party/sglang/python/sglang/jit_kernel/include/sgl_kernel/atomic.cuh
third_party/sglang/python/sglang/jit_kernel/include/sgl_kernel/cta.cuh
third_party/sglang/python/sglang/jit_kernel/include/sgl_kernel/ffi.h
third_party/sglang/python/sglang/jit_kernel/include/sgl_kernel/math.cuh
third_party/sglang/python/sglang/jit_kernel/include/sgl_kernel/runtime.cuh
third_party/sglang/python/sglang/jit_kernel/include/sgl_kernel/scalar_type.hpp
third_party/sglang/python/sglang/jit_kernel/include/sgl_kernel/source_location.h
third_party/sglang/python/sglang/jit_kernel/include/sgl_kernel/tensor.h
third_party/sglang/python/sglang/jit_kernel/include/sgl_kernel/tile.cuh
third_party/sglang/python/sglang/jit_kernel/include/sgl_kernel/type.cuh
third_party/sglang/python/sglang/jit_kernel/include/sgl_kernel/utils.cuh
third_party/sglang/python/sglang/jit_kernel/include/sgl_kernel/utils.h
third_party/sglang/python/sglang/jit_kernel/include/sgl_kernel/vec.cuh
third_party/sglang/python/sglang/jit_kernel/include/sgl_kernel/warp.cuh
third_party/sglang/python/sglang/jit_kernel/include/sgl_kernel/distributed/common.cuh
third_party/sglang/python/sglang/jit_kernel/include/sgl_kernel/distributed/custom_all_reduce.cuh
third_party/sglang/python/sglang/jit_kernel/include/sgl_kernel/impl/norm.cuh
third_party/sglang/python/sglang/jit_kernel/tests/test_add_constant.py
third_party/sglang/python/sglang/jit_kernel/tests/test_awq_dequantize.py
third_party/sglang/python/sglang/jit_kernel/tests/test_awq_marlin_moe_repack.py
third_party/sglang/python/sglang/jit_kernel/tests/test_awq_marlin_repack.py
third_party/sglang/python/sglang/jit_kernel/tests/test_cast.py
third_party/sglang/python/sglang/jit_kernel/tests/test_clamp_position.py
third_party/sglang/python/sglang/jit_kernel/tests/test_concat_mla.py
third_party/sglang/python/sglang/jit_kernel/tests/test_custom_all_reduce.py
third_party/sglang/python/sglang/jit_kernel/tests/test_cutedsl_gdn.py
third_party/sglang/python/sglang/jit_kernel/tests/test_dependency.py
third_party/sglang/python/sglang/jit_kernel/tests/test_flash_attention_4.py
third_party/sglang/python/sglang/jit_kernel/tests/test_fused_add_rmsnorm.py
third_party/sglang/python/sglang/jit_kernel/tests/test_fused_metadata_copy.py
third_party/sglang/python/sglang/jit_kernel/tests/test_fused_qknorm_rope.py
third_party/sglang/python/sglang/jit_kernel/tests/test_fused_store_index_cache.py
third_party/sglang/python/sglang/jit_kernel/tests/test_fused_verify_triton_gdn.py
third_party/sglang/python/sglang/jit_kernel/tests/test_gptq_marlin.py
third_party/sglang/python/sglang/jit_kernel/tests/test_gptq_marlin_repack.py
third_party/sglang/python/sglang/jit_kernel/tests/test_hadamard_jit.py
third_party/sglang/python/sglang/jit_kernel/tests/test_hicache.py
third_party/sglang/python/sglang/jit_kernel/tests/test_moe_align_block_size.py
third_party/sglang/python/sglang/jit_kernel/tests/test_moe_lora_align_block_size.py
third_party/sglang/python/sglang/jit_kernel/tests/test_moe_wna16_marlin.py
third_party/sglang/python/sglang/jit_kernel/tests/test_norm_tanh_mul_add_norm_scale.py
third_party/sglang/python/sglang/jit_kernel/tests/test_nvfp4_blockwise_moe.py
third_party/sglang/python/sglang/jit_kernel/tests/test_nvfp4_gemm.py
third_party/sglang/python/sglang/jit_kernel/tests/test_nvfp4_quant.py
third_party/sglang/python/sglang/jit_kernel/tests/test_per_tensor_quant_fp8.py
third_party/sglang/python/sglang/jit_kernel/tests/test_per_token_group_quant_8bit.py
third_party/sglang/python/sglang/jit_kernel/tests/test_pos_enc.py
third_party/sglang/python/sglang/jit_kernel/tests/test_qknorm.py
third_party/sglang/python/sglang/jit_kernel/tests/test_qknorm_across_heads.py
third_party/sglang/python/sglang/jit_kernel/tests/test_renorm.py
third_party/sglang/python/sglang/jit_kernel/tests/test_resolve_future_token_ids.py
third_party/sglang/python/sglang/jit_kernel/tests/test_rmsnorm.py
third_party/sglang/python/sglang/jit_kernel/tests/test_rope.py
third_party/sglang/python/sglang/jit_kernel/tests/test_store_cache.py
third_party/sglang/python/sglang/jit_kernel/tests/test_timestep_embedding.py
third_party/sglang/python/sglang/jit_kernel/tests/diffusion/test_fused_norm_scale_shift.py
third_party/sglang/python/sglang/jit_kernel/tests/diffusion/test_qknorm_rope.py
third_party/sglang/python/sglang/jit_kernel/tests/diffusion/test_qwen_image_modulation.py
third_party/sglang/python/sglang/jit_kernel/triton/gdn_fused_proj.py
third_party/sglang/python/sglang/lang/api.py
third_party/sglang/python/sglang/lang/chat_template.py
third_party/sglang/python/sglang/lang/choices.py
third_party/sglang/python/sglang/lang/interpreter.py
third_party/sglang/python/sglang/lang/ir.py
third_party/sglang/python/sglang/lang/tracer.py
third_party/sglang/python/sglang/lang/backend/anthropic.py
third_party/sglang/python/sglang/lang/backend/base_backend.py
third_party/sglang/python/sglang/lang/backend/litellm.py
third_party/sglang/python/sglang/lang/backend/openai.py
third_party/sglang/python/sglang/lang/backend/runtime_endpoint.py
third_party/sglang/python/sglang/lang/backend/vertexai.py
third_party/sglang/python/sglang/multimodal_gen/README.md
third_party/sglang/python/sglang/multimodal_gen/__init__.py
third_party/sglang/python/sglang/multimodal_gen/envs.py
third_party/sglang/python/sglang/multimodal_gen/registry.py
third_party/sglang/python/sglang/multimodal_gen/utils.py
third_party/sglang/python/sglang/multimodal_gen/.claude/CLAUDE.md
third_party/sglang/python/sglang/multimodal_gen/.claude/skills/sglang-diffusion-add-model/SKILL.md
third_party/sglang/python/sglang/multimodal_gen/.claude/skills/sglang-diffusion-ako4all-kernel/SKILL.md
third_party/sglang/python/sglang/multimodal_gen/.claude/skills/sglang-diffusion-ako4all-kernel/references/ako-loop.md
third_party/sglang/python/sglang/multimodal_gen/.claude/skills/sglang-diffusion-benchmark-profile/SKILL.md
third_party/sglang/python/sglang/multimodal_gen/.claude/skills/sglang-diffusion-benchmark-profile/benchmark-and-profile.md
third_party/sglang/python/sglang/multimodal_gen/.claude/skills/sglang-diffusion-benchmark-profile/existing-fast-paths.md
third_party/sglang/python/sglang/multimodal_gen/.claude/skills/sglang-diffusion-benchmark-profile/nsight-profiler.md
third_party/sglang/python/sglang/multimodal_gen/.claude/skills/sglang-diffusion-cuda-kernel/SKILL.md
third_party/sglang/python/sglang/multimodal_gen/.claude/skills/sglang-diffusion-cuda-kernel/references/a100-optimization-guide.md
third_party/sglang/python/sglang/multimodal_gen/.claude/skills/sglang-diffusion-cuda-kernel/references/h100-optimization-guide.md
third_party/sglang/python/sglang/multimodal_gen/.claude/skills/sglang-diffusion-cuda-kernel/references/kernel-templates.md
third_party/sglang/python/sglang/multimodal_gen/.claude/skills/sglang-diffusion-cuda-kernel/references/t4-optimization-guide.md
third_party/sglang/python/sglang/multimodal_gen/.claude/skills/sglang-diffusion-cuda-kernel/references/troubleshooting.md
third_party/sglang/python/sglang/multimodal_gen/.claude/skills/sglang-diffusion-performance/SKILL.md
third_party/sglang/python/sglang/multimodal_gen/.claude/skills/sglang-diffusion-triton-kernel/SKILL.md
third_party/sglang/python/sglang/multimodal_gen/apps/ComfyUI_SGLDiffusion/README.md
third_party/sglang/python/sglang/multimodal_gen/apps/ComfyUI_SGLDiffusion/__init__.py
third_party/sglang/python/sglang/multimodal_gen/apps/ComfyUI_SGLDiffusion/nodes.py
third_party/sglang/python/sglang/multimodal_gen/apps/ComfyUI_SGLDiffusion/utils.py
third_party/sglang/python/sglang/multimodal_gen/apps/ComfyUI_SGLDiffusion/core/__init__.py
third_party/sglang/python/sglang/multimodal_gen/apps/ComfyUI_SGLDiffusion/core/generator.py
third_party/sglang/python/sglang/multimodal_gen/apps/ComfyUI_SGLDiffusion/core/model_patcher.py
third_party/sglang/python/sglang/multimodal_gen/apps/ComfyUI_SGLDiffusion/core/server_api.py
third_party/sglang/python/sglang/multimodal_gen/apps/ComfyUI_SGLDiffusion/executors/__init__.py
third_party/sglang/python/sglang/multimodal_gen/apps/ComfyUI_SGLDiffusion/executors/base.py
third_party/sglang/python/sglang/multimodal_gen/apps/ComfyUI_SGLDiffusion/executors/flux.py
third_party/sglang/python/sglang/multimodal_gen/apps/ComfyUI_SGLDiffusion/executors/qwen_image.py
third_party/sglang/python/sglang/multimodal_gen/apps/ComfyUI_SGLDiffusion/executors/zimage.py
third_party/sglang/python/sglang/multimodal_gen/apps/ComfyUI_SGLDiffusion/test/README.md
third_party/sglang/python/sglang/multimodal_gen/apps/ComfyUI_SGLDiffusion/test/__init__.py
third_party/sglang/python/sglang/multimodal_gen/apps/ComfyUI_SGLDiffusion/test/test_flux_pipeline.py
third_party/sglang/python/sglang/multimodal_gen/apps/ComfyUI_SGLDiffusion/test/test_qwen_image_edit_pipeline.py
third_party/sglang/python/sglang/multimodal_gen/apps/ComfyUI_SGLDiffusion/test/test_qwen_image_pipeline.py
third_party/sglang/python/sglang/multimodal_gen/apps/ComfyUI_SGLDiffusion/test/test_zimage_pipeline.py
third_party/sglang/python/sglang/multimodal_gen/apps/ComfyUI_SGLDiffusion/workflows/flux_sgld_sp.json
third_party/sglang/python/sglang/multimodal_gen/apps/ComfyUI_SGLDiffusion/workflows/qwen_image_sgld.json
third_party/sglang/python/sglang/multimodal_gen/apps/ComfyUI_SGLDiffusion/workflows/sgld_image2video.json
third_party/sglang/python/sglang/multimodal_gen/apps/ComfyUI_SGLDiffusion/workflows/sgld_text2img.json
third_party/sglang/python/sglang/multimodal_gen/apps/ComfyUI_SGLDiffusion/workflows/z-image_sgld.json
third_party/sglang/python/sglang/multimodal_gen/apps/webui/README.md
third_party/sglang/python/sglang/multimodal_gen/apps/webui/__init__.py
third_party/sglang/python/sglang/multimodal_gen/apps/webui/main.py
third_party/sglang/python/sglang/multimodal_gen/benchmarks/bench_offline_throughput.py
third_party/sglang/python/sglang/multimodal_gen/benchmarks/bench_serving.py
third_party/sglang/python/sglang/multimodal_gen/benchmarks/compare_perf.py
third_party/sglang/python/sglang/multimodal_gen/benchmarks/datasets.py
third_party/sglang/python/sglang/multimodal_gen/configs/__init__.py
third_party/sglang/python/sglang/multimodal_gen/configs/utils.py
third_party/sglang/python/sglang/multimodal_gen/configs/backend/vmoba/wan_1.3B_77_448_832.json
third_party/sglang/python/sglang/multimodal_gen/configs/backend/vmoba/wan_1.3B_77_480_832.json
third_party/sglang/python/sglang/multimodal_gen/configs/models/__init__.py
third_party/sglang/python/sglang/multimodal_gen/configs/models/base.py
third_party/sglang/python/sglang/multimodal_gen/configs/models/adapter/base.py
third_party/sglang/python/sglang/multimodal_gen/configs/models/adapter/ltx_2_connector.py
third_party/sglang/python/sglang/multimodal_gen/configs/models/bridges/__init__.py
third_party/sglang/python/sglang/multimodal_gen/configs/models/bridges/mova_dual_tower.py
third_party/sglang/python/sglang/multimodal_gen/configs/models/dits/__init__.py
third_party/sglang/python/sglang/multimodal_gen/configs/models/dits/base.py
third_party/sglang/python/sglang/multimodal_gen/configs/models/dits/flux.py
third_party/sglang/python/sglang/multimodal_gen/configs/models/dits/glmimage.py
third_party/sglang/python/sglang/multimodal_gen/configs/models/dits/helios.py
third_party/sglang/python/sglang/multimodal_gen/configs/models/dits/hunyuan3d.py
third_party/sglang/python/sglang/multimodal_gen/configs/models/dits/hunyuanvideo.py
third_party/sglang/python/sglang/multimodal_gen/configs/models/dits/ltx_2.py
third_party/sglang/python/sglang/multimodal_gen/configs/models/dits/mova_audio.py
third_party/sglang/python/sglang/multimodal_gen/configs/models/dits/mova_video.py
third_party/sglang/python/sglang/multimodal_gen/configs/models/dits/qwenimage.py
third_party/sglang/python/sglang/multimodal_gen/configs/models/dits/sana.py
third_party/sglang/python/sglang/multimodal_gen/configs/models/dits/wanvideo.py
third_party/sglang/python/sglang/multimodal_gen/configs/models/dits/zimage.py
third_party/sglang/python/sglang/multimodal_gen/configs/models/encoders/__init__.py
third_party/sglang/python/sglang/multimodal_gen/configs/models/encoders/base.py
third_party/sglang/python/sglang/multimodal_gen/configs/models/encoders/clip.py
third_party/sglang/python/sglang/multimodal_gen/configs/models/encoders/gemma2.py
third_party/sglang/python/sglang/multimodal_gen/configs/models/encoders/gemma_3.py
third_party/sglang/python/sglang/multimodal_gen/configs/models/encoders/llama.py
third_party/sglang/python/sglang/multimodal_gen/configs/models/encoders/qwen3.py
third_party/sglang/python/sglang/multimodal_gen/configs/models/encoders/qwen_image.py
third_party/sglang/python/sglang/multimodal_gen/configs/models/encoders/t5.py
third_party/sglang/python/sglang/multimodal_gen/configs/models/vaes/__init__.py
third_party/sglang/python/sglang/multimodal_gen/configs/models/vaes/base.py
third_party/sglang/python/sglang/multimodal_gen/configs/models/vaes/dac.py
third_party/sglang/python/sglang/multimodal_gen/configs/models/vaes/flux.py
third_party/sglang/python/sglang/multimodal_gen/configs/models/vaes/glmimage.py
third_party/sglang/python/sglang/multimodal_gen/configs/models/vaes/hunyuan3d.py
third_party/sglang/python/sglang/multimodal_gen/configs/models/vaes/hunyuanvae.py
third_party/sglang/python/sglang/multimodal_gen/configs/models/vaes/ltx_audio.py
third_party/sglang/python/sglang/multimodal_gen/configs/models/vaes/ltx_video.py
third_party/sglang/python/sglang/multimodal_gen/configs/models/vaes/qwenimage.py
third_party/sglang/python/sglang/multimodal_gen/configs/models/vaes/sana.py
third_party/sglang/python/sglang/multimodal_gen/configs/models/vaes/wanvae.py
third_party/sglang/python/sglang/multimodal_gen/configs/models/vocoder/__init__.py
third_party/sglang/python/sglang/multimodal_gen/configs/models/vocoder/base.py
third_party/sglang/python/sglang/multimodal_gen/configs/models/vocoder/ltx_vocoder.py
third_party/sglang/python/sglang/multimodal_gen/configs/pipeline_configs/__init__.py
third_party/sglang/python/sglang/multimodal_gen/configs/pipeline_configs/base.py
third_party/sglang/python/sglang/multimodal_gen/configs/pipeline_configs/diffusers_generic.py
third_party/sglang/python/sglang/multimodal_gen/configs/pipeline_configs/flux.py
third_party/sglang/python/sglang/multimodal_gen/configs/pipeline_configs/flux_finetuned.py
third_party/sglang/python/sglang/multimodal_gen/configs/pipeline_configs/glm_image.py
third_party/sglang/python/sglang/multimodal_gen/configs/pipeline_configs/helios.py
third_party/sglang/python/sglang/multimodal_gen/configs/pipeline_configs/hunyuan.py
third_party/sglang/python/sglang/multimodal_gen/configs/pipeline_configs/hunyuan3d.py
third_party/sglang/python/sglang/multimodal_gen/configs/pipeline_configs/ltx_2.py
third_party/sglang/python/sglang/multimodal_gen/configs/pipeline_configs/mova.py
third_party/sglang/python/sglang/multimodal_gen/configs/pipeline_configs/qwen_image.py
third_party/sglang/python/sglang/multimodal_gen/configs/pipeline_configs/sana.py
third_party/sglang/python/sglang/multimodal_gen/configs/pipeline_configs/wan.py
third_party/sglang/python/sglang/multimodal_gen/configs/pipeline_configs/zimage.py
third_party/sglang/python/sglang/multimodal_gen/configs/quantization/nunchaku.py
third_party/sglang/python/sglang/multimodal_gen/configs/sample/__init__.py
third_party/sglang/python/sglang/multimodal_gen/configs/sample/diffusers_generic.py
third_party/sglang/python/sglang/multimodal_gen/configs/sample/flux.py
third_party/sglang/python/sglang/multimodal_gen/configs/sample/glmimage.py
third_party/sglang/python/sglang/multimodal_gen/configs/sample/helios.py
third_party/sglang/python/sglang/multimodal_gen/configs/sample/hunyuan.py
third_party/sglang/python/sglang/multimodal_gen/configs/sample/hunyuan3d.py
third_party/sglang/python/sglang/multimodal_gen/configs/sample/ltx_2.py
third_party/sglang/python/sglang/multimodal_gen/configs/sample/mova.py
third_party/sglang/python/sglang/multimodal_gen/configs/sample/qwenimage.py
third_party/sglang/python/sglang/multimodal_gen/configs/sample/sampling_params.py
third_party/sglang/python/sglang/multimodal_gen/configs/sample/sana.py
third_party/sglang/python/sglang/multimodal_gen/configs/sample/teacache.py
third_party/sglang/python/sglang/multimodal_gen/configs/sample/wan.py
third_party/sglang/python/sglang/multimodal_gen/configs/sample/zimage.py
third_party/sglang/python/sglang/multimodal_gen/csrc/attn/vmoba_attn/README.md
third_party/sglang/python/sglang/multimodal_gen/csrc/attn/vmoba_attn/setup.py
third_party/sglang/python/sglang/multimodal_gen/csrc/attn/vmoba_attn/tests/test_vmoba_attn.py
third_party/sglang/python/sglang/multimodal_gen/csrc/attn/vmoba_attn/vmoba/__init__.py
third_party/sglang/python/sglang/multimodal_gen/csrc/attn/vmoba_attn/vmoba/vmoba.py
third_party/sglang/python/sglang/multimodal_gen/csrc/render/__init__.py
third_party/sglang/python/sglang/multimodal_gen/csrc/render/hunyuan3d_rasterizer/__init__.py
third_party/sglang/python/sglang/multimodal_gen/csrc/render/hunyuan3d_rasterizer/rasterizer.cpp
third_party/sglang/python/sglang/multimodal_gen/csrc/render/hunyuan3d_rasterizer/rasterizer.h
third_party/sglang/python/sglang/multimodal_gen/csrc/render/hunyuan3d_rasterizer/rasterizer_gpu.cu
third_party/sglang/python/sglang/multimodal_gen/csrc/render/mesh_processor/__init__.py
third_party/sglang/python/sglang/multimodal_gen/csrc/render/mesh_processor/mesh_processor.cpp
third_party/sglang/python/sglang/multimodal_gen/runtime/launch_server.py
third_party/sglang/python/sglang/multimodal_gen/runtime/scheduler_client.py
third_party/sglang/python/sglang/multimodal_gen/runtime/server_args.py
third_party/sglang/python/sglang/multimodal_gen/runtime/cache/__init__.py
third_party/sglang/python/sglang/multimodal_gen/runtime/cache/cache_dit_integration.py
third_party/sglang/python/sglang/multimodal_gen/runtime/cache/teacache.py
third_party/sglang/python/sglang/multimodal_gen/runtime/distributed/__init__.py
third_party/sglang/python/sglang/multimodal_gen/runtime/distributed/communication_op.py
third_party/sglang/python/sglang/multimodal_gen/runtime/distributed/group_coordinator.py
third_party/sglang/python/sglang/multimodal_gen/runtime/distributed/parallel_groups.py
third_party/sglang/python/sglang/multimodal_gen/runtime/distributed/parallel_state.py
third_party/sglang/python/sglang/multimodal_gen/runtime/distributed/utils.py
third_party/sglang/python/sglang/multimodal_gen/runtime/distributed/device_communicators/__init__.py
third_party/sglang/python/sglang/multimodal_gen/runtime/distributed/device_communicators/base_device_communicator.py
third_party/sglang/python/sglang/multimodal_gen/runtime/distributed/device_communicators/cpu_communicator.py
third_party/sglang/python/sglang/multimodal_gen/runtime/distributed/device_communicators/cuda_communicator.py
third_party/sglang/python/sglang/multimodal_gen/runtime/distributed/device_communicators/pynccl.py
third_party/sglang/python/sglang/multimodal_gen/runtime/distributed/device_communicators/pynccl_wrapper.py
third_party/sglang/python/sglang/multimodal_gen/runtime/entrypoints/__init__.py
third_party/sglang/python/sglang/multimodal_gen/runtime/entrypoints/diffusion_generator.py
third_party/sglang/python/sglang/multimodal_gen/runtime/entrypoints/http_server.py
third_party/sglang/python/sglang/multimodal_gen/runtime/entrypoints/utils.py
third_party/sglang/python/sglang/multimodal_gen/runtime/entrypoints/cli/__init__.py
third_party/sglang/python/sglang/multimodal_gen/runtime/entrypoints/cli/cli_types.py
third_party/sglang/python/sglang/multimodal_gen/runtime/entrypoints/cli/generate.py
third_party/sglang/python/sglang/multimodal_gen/runtime/entrypoints/cli/main.py
third_party/sglang/python/sglang/multimodal_gen/runtime/entrypoints/cli/serve.py
third_party/sglang/python/sglang/multimodal_gen/runtime/entrypoints/cli/utils.py
third_party/sglang/python/sglang/multimodal_gen/runtime/entrypoints/openai/common_api.py
third_party/sglang/python/sglang/multimodal_gen/runtime/entrypoints/openai/image_api.py
third_party/sglang/python/sglang/multimodal_gen/runtime/entrypoints/openai/mesh_api.py
third_party/sglang/python/sglang/multimodal_gen/runtime/entrypoints/openai/protocol.py
third_party/sglang/python/sglang/multimodal_gen/runtime/entrypoints/openai/storage.py
third_party/sglang/python/sglang/multimodal_gen/runtime/entrypoints/openai/stores.py
third_party/sglang/python/sglang/multimodal_gen/runtime/entrypoints/openai/utils.py
third_party/sglang/python/sglang/multimodal_gen/runtime/entrypoints/openai/video_api.py
third_party/sglang/python/sglang/multimodal_gen/runtime/entrypoints/post_training/__init__.py
third_party/sglang/python/sglang/multimodal_gen/runtime/entrypoints/post_training/io_struct.py
third_party/sglang/python/sglang/multimodal_gen/runtime/entrypoints/post_training/weights_api.py
third_party/sglang/python/sglang/multimodal_gen/runtime/layers/__init__.py
third_party/sglang/python/sglang/multimodal_gen/runtime/layers/activation.py
third_party/sglang/python/sglang/multimodal_gen/runtime/layers/custom_op.py
third_party/sglang/python/sglang/multimodal_gen/runtime/layers/elementwise.py
third_party/sglang/python/sglang/multimodal_gen/runtime/layers/layernorm.py
third_party/sglang/python/sglang/multimodal_gen/runtime/layers/linear.py
third_party/sglang/python/sglang/multimodal_gen/runtime/layers/mlp.py
third_party/sglang/python/sglang/multimodal_gen/runtime/layers/usp.py
third_party/sglang/python/sglang/multimodal_gen/runtime/layers/utils.py
third_party/sglang/python/sglang/multimodal_gen/runtime/layers/visual_embedding.py
third_party/sglang/python/sglang/multimodal_gen/runtime/layers/vocab_parallel_embedding.py
third_party/sglang/python/sglang/multimodal_gen/runtime/layers/attention/STA_configuration.py
third_party/sglang/python/sglang/multimodal_gen/runtime/layers/attention/__init__.py
third_party/sglang/python/sglang/multimodal_gen/runtime/layers/attention/layer.py
third_party/sglang/python/sglang/multimodal_gen/runtime/layers/attention/selector.py
third_party/sglang/python/sglang/multimodal_gen/runtime/layers/attention/turbo_layer.py
third_party/sglang/python/sglang/multimodal_gen/runtime/layers/attention/backends/__init__.py
third_party/sglang/python/sglang/multimodal_gen/runtime/layers/attention/backends/aiter.py
third_party/sglang/python/sglang/multimodal_gen/runtime/layers/attention/backends/aiter_sage.py
third_party/sglang/python/sglang/multimodal_gen/runtime/layers/attention/backends/ascend_fa.py
third_party/sglang/python/sglang/multimodal_gen/runtime/layers/attention/backends/attention_backend.py
third_party/sglang/python/sglang/multimodal_gen/runtime/layers/attention/backends/flash_attn.py
third_party/sglang/python/sglang/multimodal_gen/runtime/layers/attention/backends/flash_attn_2.py
third_party/sglang/python/sglang/multimodal_gen/runtime/layers/attention/backends/sage_attn.py
third_party/sglang/python/sglang/multimodal_gen/runtime/layers/attention/backends/sage_attn3.py
third_party/sglang/python/sglang/multimodal_gen/runtime/layers/attention/backends/sdpa.py
third_party/sglang/python/sglang/multimodal_gen/runtime/layers/attention/backends/sliding_tile_attn.py
third_party/sglang/python/sglang/multimodal_gen/runtime/layers/attention/backends/sparse_linear_attn.py
third_party/sglang/python/sglang/multimodal_gen/runtime/layers/attention/backends/sparse_video_gen_2_attn.py
third_party/sglang/python/sglang/multimodal_gen/runtime/layers/attention/backends/video_sparse_attn.py
third_party/sglang/python/sglang/multimodal_gen/runtime/layers/attention/backends/vmoba.py
third_party/sglang/python/sglang/multimodal_gen/runtime/layers/lora/linear.py
third_party/sglang/python/sglang/multimodal_gen/runtime/layers/quantization/__init__.py
third_party/sglang/python/sglang/multimodal_gen/runtime/layers/quantization/fp8.py
third_party/sglang/python/sglang/multimodal_gen/runtime/layers/quantization/modelopt_quant.py
third_party/sglang/python/sglang/multimodal_gen/runtime/layers/quantization/modelslim.py
third_party/sglang/python/sglang/multimodal_gen/runtime/layers/quantization/nunchaku_linear.py
third_party/sglang/python/sglang/multimodal_gen/runtime/layers/quantization/configs/base_config.py
third_party/sglang/python/sglang/multimodal_gen/runtime/layers/quantization/configs/nunchaku_config.py
third_party/sglang/python/sglang/multimodal_gen/runtime/layers/rotary_embedding/__init__.py
third_party/sglang/python/sglang/multimodal_gen/runtime/layers/rotary_embedding/base.py
third_party/sglang/python/sglang/multimodal_gen/runtime/layers/rotary_embedding/factory.py
third_party/sglang/python/sglang/multimodal_gen/runtime/layers/rotary_embedding/mrope.py
third_party/sglang/python/sglang/multimodal_gen/runtime/layers/rotary_embedding/utils.py
third_party/sglang/python/sglang/multimodal_gen/runtime/loader/fsdp_load.py
third_party/sglang/python/sglang/multimodal_gen/runtime/loader/transformer_load_utils.py
third_party/sglang/python/sglang/multimodal_gen/runtime/loader/utils.py
third_party/sglang/python/sglang/multimodal_gen/runtime/loader/weight_utils.py
third_party/sglang/python/sglang/multimodal_gen/runtime/loader/weights_updater.py
third_party/sglang/python/sglang/multimodal_gen/runtime/loader/component_loaders/adapter_loader.py
third_party/sglang/python/sglang/multimodal_gen/runtime/loader/component_loaders/bridge_loader.py
third_party/sglang/python/sglang/multimodal_gen/runtime/loader/component_loaders/component_loader.py
third_party/sglang/python/sglang/multimodal_gen/runtime/loader/component_loaders/image_encoder_loader.py
third_party/sglang/python/sglang/multimodal_gen/runtime/loader/component_loaders/scheduler_loader.py
third_party/sglang/python/sglang/multimodal_gen/runtime/loader/component_loaders/text_encoder_loader.py
third_party/sglang/python/sglang/multimodal_gen/runtime/loader/component_loaders/transformer_loader.py
third_party/sglang/python/sglang/multimodal_gen/runtime/loader/component_loaders/upsampler_loader.py
third_party/sglang/python/sglang/multimodal_gen/runtime/loader/component_loaders/vae_loader.py
third_party/sglang/python/sglang/multimodal_gen/runtime/loader/component_loaders/vl_encoder_loader.py
third_party/sglang/python/sglang/multimodal_gen/runtime/loader/component_loaders/vocoder_loader.py
third_party/sglang/python/sglang/multimodal_gen/runtime/managers/forward_context.py
third_party/sglang/python/sglang/multimodal_gen/runtime/managers/gpu_worker.py
third_party/sglang/python/sglang/multimodal_gen/runtime/managers/scheduler.py
third_party/sglang/python/sglang/multimodal_gen/runtime/models/__init__.py
third_party/sglang/python/sglang/multimodal_gen/runtime/models/parameter.py
third_party/sglang/python/sglang/multimodal_gen/runtime/models/registry.py
third_party/sglang/python/sglang/multimodal_gen/runtime/models/utils.py
third_party/sglang/python/sglang/multimodal_gen/runtime/models/vision_utils.py
third_party/sglang/python/sglang/multimodal_gen/runtime/models/adapter/ltx_2_connector.py
third_party/sglang/python/sglang/multimodal_gen/runtime/models/bridges/__init__.py
third_party/sglang/python/sglang/multimodal_gen/runtime/models/bridges/mova_dual_tower.py
third_party/sglang/python/sglang/multimodal_gen/runtime/models/dits/base.py
third_party/sglang/python/sglang/multimodal_gen/runtime/models/dits/causal_wanvideo.py
third_party/sglang/python/sglang/multimodal_gen/runtime/models/dits/flux.py
third_party/sglang/python/sglang/multimodal_gen/runtime/models/dits/flux_2.py
third_party/sglang/python/sglang/multimodal_gen/runtime/models/dits/glm_image.py
third_party/sglang/python/sglang/multimodal_gen/runtime/models/dits/helios.py
third_party/sglang/python/sglang/multimodal_gen/runtime/models/dits/hunyuan3d.py
third_party/sglang/python/sglang/multimodal_gen/runtime/models/dits/hunyuanvideo.py
third_party/sglang/python/sglang/multimodal_gen/runtime/models/dits/ltx_2.py
third_party/sglang/python/sglang/multimodal_gen/runtime/models/dits/mova_audio_dit.py
third_party/sglang/python/sglang/multimodal_gen/runtime/models/dits/mova_video_dit.py
third_party/sglang/python/sglang/multimodal_gen/runtime/models/dits/qwen_image.py
third_party/sglang/python/sglang/multimodal_gen/runtime/models/dits/sana.py
third_party/sglang/python/sglang/multimodal_gen/runtime/models/dits/wanvideo.py
third_party/sglang/python/sglang/multimodal_gen/runtime/models/dits/zimage.py
third_party/sglang/python/sglang/multimodal_gen/runtime/models/encoders/base.py
third_party/sglang/python/sglang/multimodal_gen/runtime/models/encoders/bert.py
third_party/sglang/python/sglang/multimodal_gen/runtime/models/encoders/clip.py
third_party/sglang/python/sglang/multimodal_gen/runtime/models/encoders/gemma2.py
third_party/sglang/python/sglang/multimodal_gen/runtime/models/encoders/gemma_3.py
third_party/sglang/python/sglang/multimodal_gen/runtime/models/encoders/hunyuan3d.py
third_party/sglang/python/sglang/multimodal_gen/runtime/models/encoders/llama.py
third_party/sglang/python/sglang/multimodal_gen/runtime/models/encoders/mistral_3.py
third_party/sglang/python/sglang/multimodal_gen/runtime/models/encoders/qwen2_5vl.py
third_party/sglang/python/sglang/multimodal_gen/runtime/models/encoders/qwen3.py
third_party/sglang/python/sglang/multimodal_gen/runtime/models/encoders/t5.py
third_party/sglang/python/sglang/multimodal_gen/runtime/models/encoders/vision.py
third_party/sglang/python/sglang/multimodal_gen/runtime/models/schedulers/base.py
third_party/sglang/python/sglang/multimodal_gen/runtime/models/schedulers/flow_match_pair.py
third_party/sglang/python/sglang/multimodal_gen/runtime/models/schedulers/hunyuan3d_scheduler.py
third_party/sglang/python/sglang/multimodal_gen/runtime/models/schedulers/scheduling_comfyui_passthrough.py
third_party/sglang/python/sglang/multimodal_gen/runtime/models/schedulers/scheduling_dpm_solver_multistep.py
third_party/sglang/python/sglang/multimodal_gen/runtime/models/schedulers/scheduling_flow_match_euler_discrete.py
third_party/sglang/python/sglang/multimodal_gen/runtime/models/schedulers/scheduling_flow_unipc_multistep.py
third_party/sglang/python/sglang/multimodal_gen/runtime/models/schedulers/scheduling_helios.py
third_party/sglang/python/sglang/multimodal_gen/runtime/models/schedulers/scheduling_self_forcing_flow_match.py
third_party/sglang/python/sglang/multimodal_gen/runtime/models/schedulers/scheduling_unipc_multistep.py
third_party/sglang/python/sglang/multimodal_gen/runtime/models/upsampler/__init__.py
third_party/sglang/python/sglang/multimodal_gen/runtime/models/upsampler/latent_upsampler.py
third_party/sglang/python/sglang/multimodal_gen/runtime/models/vaes/autoencoder.py
third_party/sglang/python/sglang/multimodal_gen/runtime/models/vaes/autoencoder_dc.py
third_party/sglang/python/sglang/multimodal_gen/runtime/models/vaes/autoencoder_kl_flux2.py
third_party/sglang/python/sglang/multimodal_gen/runtime/models/vaes/autoencoder_kl_qwenimage.py
third_party/sglang/python/sglang/multimodal_gen/runtime/models/vaes/common.py
third_party/sglang/python/sglang/multimodal_gen/runtime/models/vaes/dac.py
third_party/sglang/python/sglang/multimodal_gen/runtime/models/vaes/hunyuan3d_vae.py
third_party/sglang/python/sglang/multimodal_gen/runtime/models/vaes/hunyuanvae.py
third_party/sglang/python/sglang/multimodal_gen/runtime/models/vaes/ltx_2_audio.py
third_party/sglang/python/sglang/multimodal_gen/runtime/models/vaes/ltx_2_vae.py
third_party/sglang/python/sglang/multimodal_gen/runtime/models/vaes/wanvae.py
third_party/sglang/python/sglang/multimodal_gen/runtime/models/vaes/parallel/wan_common_utils.py
third_party/sglang/python/sglang/multimodal_gen/runtime/models/vaes/parallel/wan_dist_utils.py
third_party/sglang/python/sglang/multimodal_gen/runtime/models/vocoder/ltx_2_vocoder.py
third_party/sglang/python/sglang/multimodal_gen/runtime/pipelines/__init__.py
third_party/sglang/python/sglang/multimodal_gen/runtime/pipelines/comfyui_flux_pipeline.py
third_party/sglang/python/sglang/multimodal_gen/runtime/pipelines/comfyui_qwen_image_pipeline.py
third_party/sglang/python/sglang/multimodal_gen/runtime/pipelines/comfyui_zimage_pipeline.py
third_party/sglang/python/sglang/multimodal_gen/runtime/pipelines/diffusers_pipeline.py
third_party/sglang/python/sglang/multimodal_gen/runtime/pipelines/flux.py
third_party/sglang/python/sglang/multimodal_gen/runtime/pipelines/flux_2.py
third_party/sglang/python/sglang/multimodal_gen/runtime/pipelines/flux_2_klein.py
third_party/sglang/python/sglang/multimodal_gen/runtime/pipelines/flux_2_nvfp4.py
third_party/sglang/python/sglang/multimodal_gen/runtime/pipelines/glm_image.py
third_party/sglang/python/sglang/multimodal_gen/runtime/pipelines/helios_pipeline.py
third_party/sglang/python/sglang/multimodal_gen/runtime/pipelines/hunyuan3d_pipeline.py
third_party/sglang/python/sglang/multimodal_gen/runtime/pipelines/hunyuan_pipeline.py
third_party/sglang/python/sglang/multimodal_gen/runtime/pipelines/ltx_2_pipeline.py
third_party/sglang/python/sglang/multimodal_gen/runtime/pipelines/mova_pipeline.py
third_party/sglang/python/sglang/multimodal_gen/runtime/pipelines/qwen_image.py
third_party/sglang/python/sglang/multimodal_gen/runtime/pipelines/sana.py
third_party/sglang/python/sglang/multimodal_gen/runtime/pipelines/wan_causal_dmd_pipeline.py
third_party/sglang/python/sglang/multimodal_gen/runtime/pipelines/wan_dmd_pipeline.py
third_party/sglang/python/sglang/multimodal_gen/runtime/pipelines/wan_i2v_dmd_pipeline.py
third_party/sglang/python/sglang/multimodal_gen/runtime/pipelines/wan_i2v_pipeline.py
third_party/sglang/python/sglang/multimodal_gen/runtime/pipelines/wan_pipeline.py
third_party/sglang/python/sglang/multimodal_gen/runtime/pipelines/zimage_pipeline.py
third_party/sglang/python/sglang/multimodal_gen/runtime/pipelines_core/__init__.py
third_party/sglang/python/sglang/multimodal_gen/runtime/pipelines_core/composed_pipeline_base.py
third_party/sglang/python/sglang/multimodal_gen/runtime/pipelines_core/lora_format_adapter.py
third_party/sglang/python/sglang/multimodal_gen/runtime/pipelines_core/lora_pipeline.py
third_party/sglang/python/sglang/multimodal_gen/runtime/pipelines_core/schedule_batch.py
third_party/sglang/python/sglang/multimodal_gen/runtime/pipelines_core/executors/parallel_executor.py
third_party/sglang/python/sglang/multimodal_gen/runtime/pipelines_core/executors/pipeline_executor.py
third_party/sglang/python/sglang/multimodal_gen/runtime/pipelines_core/executors/sync_executor.py
third_party/sglang/python/sglang/multimodal_gen/runtime/pipelines_core/stages/__init__.py
third_party/sglang/python/sglang/multimodal_gen/runtime/pipelines_core/stages/base.py
third_party/sglang/python/sglang/multimodal_gen/runtime/pipelines_core/stages/causal_denoising.py
third_party/sglang/python/sglang/multimodal_gen/runtime/pipelines_core/stages/comfyui_latent_preparation.py
third_party/sglang/python/sglang/multimodal_gen/runtime/pipelines_core/stages/decoding.py
third_party/sglang/python/sglang/multimodal_gen/runtime/pipelines_core/stages/decoding_av.py
third_party/sglang/python/sglang/multimodal_gen/runtime/pipelines_core/stages/denoising.py
third_party/sglang/python/sglang/multimodal_gen/runtime/pipelines_core/stages/denoising_av.py
third_party/sglang/python/sglang/multimodal_gen/runtime/pipelines_core/stages/denoising_dmd.py
third_party/sglang/python/sglang/multimodal_gen/runtime/pipelines_core/stages/encoding.py
third_party/sglang/python/sglang/multimodal_gen/runtime/pipelines_core/stages/hunyuan3d_paint.py
third_party/sglang/python/sglang/multimodal_gen/runtime/pipelines_core/stages/hunyuan3d_shape.py
third_party/sglang/python/sglang/multimodal_gen/runtime/pipelines_core/stages/image_encoding.py
third_party/sglang/python/sglang/multimodal_gen/runtime/pipelines_core/stages/input_validation.py
third_party/sglang/python/sglang/multimodal_gen/runtime/pipelines_core/stages/latent_preparation.py
third_party/sglang/python/sglang/multimodal_gen/runtime/pipelines_core/stages/latent_preparation_av.py
third_party/sglang/python/sglang/multimodal_gen/runtime/pipelines_core/stages/text_connector.py
third_party/sglang/python/sglang/multimodal_gen/runtime/pipelines_core/stages/text_encoding.py
third_party/sglang/python/sglang/multimodal_gen/runtime/pipelines_core/stages/timestep_preparation.py
third_party/sglang/python/sglang/multimodal_gen/runtime/pipelines_core/stages/upsampling.py
third_party/sglang/python/sglang/multimodal_gen/runtime/pipelines_core/stages/validators.py
third_party/sglang/python/sglang/multimodal_gen/runtime/pipelines_core/stages/model_specific_stages/glm_image.py
third_party/sglang/python/sglang/multimodal_gen/runtime/pipelines_core/stages/model_specific_stages/helios_decoding.py
third_party/sglang/python/sglang/multimodal_gen/runtime/pipelines_core/stages/model_specific_stages/helios_denoising.py
third_party/sglang/python/sglang/multimodal_gen/runtime/pipelines_core/stages/model_specific_stages/mova.py
third_party/sglang/python/sglang/multimodal_gen/runtime/pipelines_core/stages/model_specific_stages/qwen_image_layered.py
third_party/sglang/python/sglang/multimodal_gen/runtime/platforms/__init__.py
third_party/sglang/python/sglang/multimodal_gen/runtime/platforms/cpu.py
third_party/sglang/python/sglang/multimodal_gen/runtime/platforms/cuda.py
third_party/sglang/python/sglang/multimodal_gen/runtime/platforms/interface.py
third_party/sglang/python/sglang/multimodal_gen/runtime/platforms/mps.py
third_party/sglang/python/sglang/multimodal_gen/runtime/platforms/musa.py
third_party/sglang/python/sglang/multimodal_gen/runtime/platforms/npu.py
third_party/sglang/python/sglang/multimodal_gen/runtime/platforms/rocm.py
third_party/sglang/python/sglang/multimodal_gen/runtime/postprocess/__init__.py
third_party/sglang/python/sglang/multimodal_gen/runtime/postprocess/realesrgan_upscaler.py
third_party/sglang/python/sglang/multimodal_gen/runtime/postprocess/rife_interpolator.py
third_party/sglang/python/sglang/multimodal_gen/runtime/utils/common.py
third_party/sglang/python/sglang/multimodal_gen/runtime/utils/distributed.py
third_party/sglang/python/sglang/multimodal_gen/runtime/utils/hf_diffusers_utils.py
third_party/sglang/python/sglang/multimodal_gen/runtime/utils/layerwise_offload.py
third_party/sglang/python/sglang/multimodal_gen/runtime/utils/logging_utils.py
third_party/sglang/python/sglang/multimodal_gen/runtime/utils/mesh3d_utils.py
third_party/sglang/python/sglang/multimodal_gen/runtime/utils/model_overlay.py
third_party/sglang/python/sglang/multimodal_gen/runtime/utils/perf_logger.py
third_party/sglang/python/sglang/multimodal_gen/runtime/utils/profiler.py
third_party/sglang/python/sglang/multimodal_gen/runtime/utils/quantization_utils.py
third_party/sglang/python/sglang/multimodal_gen/test/__init__.py
third_party/sglang/python/sglang/multimodal_gen/test/run_suite.py
third_party/sglang/python/sglang/multimodal_gen/test/slack_utils.py
third_party/sglang/python/sglang/multimodal_gen/test/test_utils.py
third_party/sglang/python/sglang/multimodal_gen/test/cli/test_generate_common.py
third_party/sglang/python/sglang/multimodal_gen/test/cli/test_generate_i2i.py
third_party/sglang/python/sglang/multimodal_gen/test/cli/test_generate_t2i_perf.py
third_party/sglang/python/sglang/multimodal_gen/test/manual/test_diffusion_srt_fp4_linear.py
third_party/sglang/python/sglang/multimodal_gen/test/scripts/gen_diffusion_ci_outputs.py
third_party/sglang/python/sglang/multimodal_gen/test/scripts/gen_perf_baselines.py
third_party/sglang/python/sglang/multimodal_gen/test/server/accuracy_config.py
third_party/sglang/python/sglang/multimodal_gen/test/server/accuracy_hooks.py
third_party/sglang/python/sglang/multimodal_gen/test/server/accuracy_utils.py
third_party/sglang/python/sglang/multimodal_gen/test/server/component_accuracy.py
third_party/sglang/python/sglang/multimodal_gen/test/server/conftest.py
third_party/sglang/python/sglang/multimodal_gen/test/server/perf_baselines.json
third_party/sglang/python/sglang/multimodal_gen/test/server/test_accuracy_1_gpu_a.py
third_party/sglang/python/sglang/multimodal_gen/test/server/test_accuracy_1_gpu_b.py
third_party/sglang/python/sglang/multimodal_gen/test/server/test_accuracy_2_gpu_a.py
third_party/sglang/python/sglang/multimodal_gen/test/server/test_accuracy_2_gpu_b.py
third_party/sglang/python/sglang/multimodal_gen/test/server/test_server_2_gpu_a.py
third_party/sglang/python/sglang/multimodal_gen/test/server/test_server_2_gpu_b.py
third_party/sglang/python/sglang/multimodal_gen/test/server/test_server_a.py
third_party/sglang/python/sglang/multimodal_gen/test/server/test_server_b.py
third_party/sglang/python/sglang/multimodal_gen/test/server/test_server_c.py
third_party/sglang/python/sglang/multimodal_gen/test/server/test_server_common.py
third_party/sglang/python/sglang/multimodal_gen/test/server/test_server_utils.py
third_party/sglang/python/sglang/multimodal_gen/test/server/test_update_weights_from_disk.py
third_party/sglang/python/sglang/multimodal_gen/test/server/testcase_configs.py
third_party/sglang/python/sglang/multimodal_gen/test/server/ascend/perf_baselines_npu.json
third_party/sglang/python/sglang/multimodal_gen/test/server/ascend/test_server_1_npu.py
third_party/sglang/python/sglang/multimodal_gen/test/server/ascend/test_server_2_npu.py
third_party/sglang/python/sglang/multimodal_gen/test/server/ascend/test_server_8_npu.py
third_party/sglang/python/sglang/multimodal_gen/test/server/ascend/testcase_configs_npu.py
third_party/sglang/python/sglang/multimodal_gen/test/test_files/launch_flux.json
third_party/sglang/python/sglang/multimodal_gen/test/test_files/launch_wan.json
third_party/sglang/python/sglang/multimodal_gen/test/unit/test_input_validation.py
third_party/sglang/python/sglang/multimodal_gen/test/unit/test_lora_format_adapter.py
third_party/sglang/python/sglang/multimodal_gen/test/unit/test_resolve_prompts.py
third_party/sglang/python/sglang/multimodal_gen/test/unit/test_sampling_params.py
third_party/sglang/python/sglang/multimodal_gen/test/unit/test_server_args.py
third_party/sglang/python/sglang/multimodal_gen/test/unit/test_storage.py
third_party/sglang/python/sglang/multimodal_gen/third_party/__init__.py
third_party/sglang/python/sglang/multimodal_gen/third_party/pynvml.py
third_party/sglang/python/sglang/multimodal_gen/tools/convert_hf_to_fp8.py
third_party/sglang/python/sglang/multimodal_gen/tools/wan_repack.py
third_party/sglang/python/sglang/srt/constants.py
third_party/sglang/python/sglang/srt/environ.py
third_party/sglang/python/sglang/srt/server_args.py
third_party/sglang/python/sglang/srt/server_args_config_parser.py
third_party/sglang/python/sglang/srt/batch_invariant_ops/__init__.py
third_party/sglang/python/sglang/srt/batch_invariant_ops/batch_invariant_ops.py
third_party/sglang/python/sglang/srt/batch_overlap/operations.py
third_party/sglang/python/sglang/srt/batch_overlap/operations_strategy.py
third_party/sglang/python/sglang/srt/batch_overlap/single_batch_overlap.py
third_party/sglang/python/sglang/srt/batch_overlap/two_batch_overlap.py
third_party/sglang/python/sglang/srt/checkpoint_engine/__init__.py
third_party/sglang/python/sglang/srt/checkpoint_engine/checkpoint_engine_worker.py
third_party/sglang/python/sglang/srt/checkpoint_engine/update.py
third_party/sglang/python/sglang/srt/compilation/backend.py
third_party/sglang/python/sglang/srt/compilation/compilation_config.py
third_party/sglang/python/sglang/srt/compilation/compilation_counter.py
third_party/sglang/python/sglang/srt/compilation/compile.py
third_party/sglang/python/sglang/srt/compilation/compiler_interface.py
third_party/sglang/python/sglang/srt/compilation/cuda_piecewise_backend.py
third_party/sglang/python/sglang/srt/compilation/fix_functionalization.py
third_party/sglang/python/sglang/srt/compilation/fx_utils.py
third_party/sglang/python/sglang/srt/compilation/inductor_pass.py
third_party/sglang/python/sglang/srt/compilation/npu_piecewise_backend.py
third_party/sglang/python/sglang/srt/compilation/pass_manager.py
third_party/sglang/python/sglang/srt/compilation/piecewise_context_manager.py
third_party/sglang/python/sglang/srt/compilation/weak_ref_tensor.py
third_party/sglang/python/sglang/srt/configs/__init__.py
third_party/sglang/python/sglang/srt/configs/afmoe.py
third_party/sglang/python/sglang/srt/configs/bailing_hybrid.py
third_party/sglang/python/sglang/srt/configs/chatglm.py
third_party/sglang/python/sglang/srt/configs/dbrx.py
third_party/sglang/python/sglang/srt/configs/deepseek_ocr.py
third_party/sglang/python/sglang/srt/configs/deepseekvl2.py
third_party/sglang/python/sglang/srt/configs/device_config.py
third_party/sglang/python/sglang/srt/configs/dots_ocr.py
third_party/sglang/python/sglang/srt/configs/dots_vlm.py
third_party/sglang/python/sglang/srt/configs/exaone.py
third_party/sglang/python/sglang/srt/configs/falcon_h1.py
third_party/sglang/python/sglang/srt/configs/granitemoehybrid.py
third_party/sglang/python/sglang/srt/configs/internvl.py
third_party/sglang/python/sglang/srt/configs/janus_pro.py
third_party/sglang/python/sglang/srt/configs/jet_nemotron.py
third_party/sglang/python/sglang/srt/configs/jet_vlm.py
third_party/sglang/python/sglang/srt/configs/kimi_k25.py
third_party/sglang/python/sglang/srt/configs/kimi_linear.py
third_party/sglang/python/sglang/srt/configs/kimi_vl.py
third_party/sglang/python/sglang/srt/configs/kimi_vl_moonvit.py
third_party/sglang/python/sglang/srt/configs/lfm2.py
third_party/sglang/python/sglang/srt/configs/lfm2_moe.py
third_party/sglang/python/sglang/srt/configs/lfm2_vl.py
third_party/sglang/python/sglang/srt/configs/load_config.py
third_party/sglang/python/sglang/srt/configs/longcat_flash.py
third_party/sglang/python/sglang/srt/configs/mamba_utils.py
third_party/sglang/python/sglang/srt/configs/model_config.py
third_party/sglang/python/sglang/srt/configs/modelopt_config.py
third_party/sglang/python/sglang/srt/configs/nano_nemotron_vl.py
third_party/sglang/python/sglang/srt/configs/nemotron_h.py
third_party/sglang/python/sglang/srt/configs/olmo3.py
third_party/sglang/python/sglang/srt/configs/points_v15_chat.py
third_party/sglang/python/sglang/srt/configs/qwen3_5.py
third_party/sglang/python/sglang/srt/configs/qwen3_next.py
third_party/sglang/python/sglang/srt/configs/qwen3_omni.py
third_party/sglang/python/sglang/srt/configs/qwen3_vl.py
third_party/sglang/python/sglang/srt/configs/radio.py
third_party/sglang/python/sglang/srt/configs/step3_vl.py
third_party/sglang/python/sglang/srt/configs/step3p5.py
third_party/sglang/python/sglang/srt/configs/update_config.py
third_party/sglang/python/sglang/srt/configs/utils.py
third_party/sglang/python/sglang/srt/connector/__init__.py
third_party/sglang/python/sglang/srt/connector/base_connector.py
third_party/sglang/python/sglang/srt/connector/redis.py
third_party/sglang/python/sglang/srt/connector/remote_instance.py
third_party/sglang/python/sglang/srt/connector/s3.py
third_party/sglang/python/sglang/srt/connector/utils.py
third_party/sglang/python/sglang/srt/connector/serde/__init__.py
third_party/sglang/python/sglang/srt/connector/serde/safe_serde.py
third_party/sglang/python/sglang/srt/connector/serde/serde.py
third_party/sglang/python/sglang/srt/constrained/base_grammar_backend.py
third_party/sglang/python/sglang/srt/constrained/grammar_manager.py
third_party/sglang/python/sglang/srt/constrained/llguidance_backend.py
third_party/sglang/python/sglang/srt/constrained/outlines_backend.py
third_party/sglang/python/sglang/srt/constrained/outlines_jump_forward.py
third_party/sglang/python/sglang/srt/constrained/reasoner_grammar_backend.py
third_party/sglang/python/sglang/srt/constrained/utils.py
third_party/sglang/python/sglang/srt/constrained/xgrammar_backend.py
third_party/sglang/python/sglang/srt/constrained/triton_ops/bitmask_ops.py
third_party/sglang/python/sglang/srt/debug_utils/__init__.py
third_party/sglang/python/sglang/srt/debug_utils/cuda_coredump.py
third_party/sglang/python/sglang/srt/debug_utils/dump_comparator.py
third_party/sglang/python/sglang/srt/debug_utils/dump_loader.py
third_party/sglang/python/sglang/srt/debug_utils/dumper.py
third_party/sglang/python/sglang/srt/debug_utils/log_parser.py
third_party/sglang/python/sglang/srt/debug_utils/model_truncator.py
third_party/sglang/python/sglang/srt/debug_utils/tensor_dump_forward_hook.py
third_party/sglang/python/sglang/srt/debug_utils/text_comparator.py
third_party/sglang/python/sglang/srt/debug_utils/comparator/__init__.py
third_party/sglang/python/sglang/srt/debug_utils/comparator/__main__.py
third_party/sglang/python/sglang/srt/debug_utils/comparator/bundle_comparator.py
third_party/sglang/python/sglang/srt/debug_utils/comparator/bundle_matcher.py
third_party/sglang/python/sglang/srt/debug_utils/comparator/display.py
third_party/sglang/python/sglang/srt/debug_utils/comparator/dp_utils.py
third_party/sglang/python/sglang/srt/debug_utils/comparator/entrypoint.py
third_party/sglang/python/sglang/srt/debug_utils/comparator/log_sink.py
third_party/sglang/python/sglang/srt/debug_utils/comparator/meta_overrider.py
third_party/sglang/python/sglang/srt/debug_utils/comparator/output_formatter.py
third_party/sglang/python/sglang/srt/debug_utils/comparator/output_types.py
third_party/sglang/python/sglang/srt/debug_utils/comparator/per_token_visualizer.py
third_party/sglang/python/sglang/srt/debug_utils/comparator/preset.py
third_party/sglang/python/sglang/srt/debug_utils/comparator/report_sink.py
third_party/sglang/python/sglang/srt/debug_utils/comparator/utils.py
third_party/sglang/python/sglang/srt/debug_utils/comparator/aligner/__init__.py
third_party/sglang/python/sglang/srt/debug_utils/comparator/aligner/axis_aligner.py
third_party/sglang/python/sglang/srt/debug_utils/comparator/aligner/entrypoint/__init__.py
third_party/sglang/python/sglang/srt/debug_utils/comparator/aligner/entrypoint/executor.py
third_party/sglang/python/sglang/srt/debug_utils/comparator/aligner/entrypoint/planner.py
third_party/sglang/python/sglang/srt/debug_utils/comparator/aligner/entrypoint/traced_types.py
third_party/sglang/python/sglang/srt/debug_utils/comparator/aligner/entrypoint/types.py
third_party/sglang/python/sglang/srt/debug_utils/comparator/aligner/reorderer/__init__.py
third_party/sglang/python/sglang/srt/debug_utils/comparator/aligner/reorderer/executor.py
third_party/sglang/python/sglang/srt/debug_utils/comparator/aligner/reorderer/planner.py
third_party/sglang/python/sglang/srt/debug_utils/comparator/aligner/reorderer/types.py
third_party/sglang/python/sglang/srt/debug_utils/comparator/aligner/token_aligner/__init__.py
third_party/sglang/python/sglang/srt/debug_utils/comparator/aligner/token_aligner/entrypoint.py
third_party/sglang/python/sglang/srt/debug_utils/comparator/aligner/token_aligner/concat_steps/__init__.py
third_party/sglang/python/sglang/srt/debug_utils/comparator/aligner/token_aligner/concat_steps/executor.py
third_party/sglang/python/sglang/srt/debug_utils/comparator/aligner/token_aligner/concat_steps/thd_seq_lens_loader.py
third_party/sglang/python/sglang/srt/debug_utils/comparator/aligner/token_aligner/smart/__init__.py
third_party/sglang/python/sglang/srt/debug_utils/comparator/aligner/token_aligner/smart/aux_loader.py
third_party/sglang/python/sglang/srt/debug_utils/comparator/aligner/token_aligner/smart/aux_plugins.py
third_party/sglang/python/sglang/srt/debug_utils/comparator/aligner/token_aligner/smart/executor.py
third_party/sglang/python/sglang/srt/debug_utils/comparator/aligner/token_aligner/smart/planner.py
third_party/sglang/python/sglang/srt/debug_utils/comparator/aligner/token_aligner/smart/seq_info_builder.py
third_party/sglang/python/sglang/srt/debug_utils/comparator/aligner/token_aligner/smart/types.py
third_party/sglang/python/sglang/srt/debug_utils/comparator/aligner/unsharder/__init__.py
third_party/sglang/python/sglang/srt/debug_utils/comparator/aligner/unsharder/executor.py
third_party/sglang/python/sglang/srt/debug_utils/comparator/aligner/unsharder/parallel_info.py
third_party/sglang/python/sglang/srt/debug_utils/comparator/aligner/unsharder/planner.py
third_party/sglang/python/sglang/srt/debug_utils/comparator/aligner/unsharder/types.py
third_party/sglang/python/sglang/srt/debug_utils/comparator/dims_spec/__init__.py
third_party/sglang/python/sglang/srt/debug_utils/comparator/dims_spec/comment_parser.py
third_party/sglang/python/sglang/srt/debug_utils/comparator/dims_spec/dim_parser.py
third_party/sglang/python/sglang/srt/debug_utils/comparator/dims_spec/dims_parser.py
third_party/sglang/python/sglang/srt/debug_utils/comparator/dims_spec/modifier_parser.py
third_party/sglang/python/sglang/srt/debug_utils/comparator/dims_spec/tensor_naming.py
third_party/sglang/python/sglang/srt/debug_utils/comparator/dims_spec/types.py
third_party/sglang/python/sglang/srt/debug_utils/comparator/tensor_comparator/__init__.py
third_party/sglang/python/sglang/srt/debug_utils/comparator/tensor_comparator/comparator.py
third_party/sglang/python/sglang/srt/debug_utils/comparator/tensor_comparator/formatter.py
third_party/sglang/python/sglang/srt/debug_utils/comparator/tensor_comparator/types.py
third_party/sglang/python/sglang/srt/debug_utils/comparator/visualizer/__init__.py
third_party/sglang/python/sglang/srt/debug_utils/comparator/visualizer/figure.py
third_party/sglang/python/sglang/srt/debug_utils/comparator/visualizer/panels.py
third_party/sglang/python/sglang/srt/debug_utils/comparator/visualizer/preprocessing.py
third_party/sglang/python/sglang/srt/debug_utils/schedule_simulator/__init__.py
third_party/sglang/python/sglang/srt/debug_utils/schedule_simulator/__main__.py
third_party/sglang/python/sglang/srt/debug_utils/schedule_simulator/entrypoint.py
third_party/sglang/python/sglang/srt/debug_utils/schedule_simulator/gpu_state.py
third_party/sglang/python/sglang/srt/debug_utils/schedule_simulator/metrics.py
third_party/sglang/python/sglang/srt/debug_utils/schedule_simulator/request.py
third_party/sglang/python/sglang/srt/debug_utils/schedule_simulator/simulator.py
third_party/sglang/python/sglang/srt/debug_utils/schedule_simulator/data_source/__init__.py
third_party/sglang/python/sglang/srt/debug_utils/schedule_simulator/data_source/data_loader.py
third_party/sglang/python/sglang/srt/debug_utils/schedule_simulator/data_source/data_synthesis.py
third_party/sglang/python/sglang/srt/debug_utils/schedule_simulator/routers/__init__.py
third_party/sglang/python/sglang/srt/debug_utils/schedule_simulator/routers/base.py
third_party/sglang/python/sglang/srt/debug_utils/schedule_simulator/routers/random_router.py
third_party/sglang/python/sglang/srt/debug_utils/schedule_simulator/routers/round_robin_router.py
third_party/sglang/python/sglang/srt/debug_utils/schedule_simulator/routers/sticky_router.py
third_party/sglang/python/sglang/srt/debug_utils/schedule_simulator/schedulers/__init__.py
third_party/sglang/python/sglang/srt/debug_utils/schedule_simulator/schedulers/base.py
third_party/sglang/python/sglang/srt/debug_utils/schedule_simulator/schedulers/fifo_scheduler.py
third_party/sglang/python/sglang/srt/debug_utils/source_patcher/__init__.py
third_party/sglang/python/sglang/srt/debug_utils/source_patcher/code_patcher.py
third_party/sglang/python/sglang/srt/debug_utils/source_patcher/source_editor.py
third_party/sglang/python/sglang/srt/debug_utils/source_patcher/types.py
third_party/sglang/python/sglang/srt/disaggregation/decode.py
third_party/sglang/python/sglang/srt/disaggregation/decode_kvcache_offload_manager.py
third_party/sglang/python/sglang/srt/disaggregation/decode_schedule_batch_mixin.py
third_party/sglang/python/sglang/srt/disaggregation/encode_grpc_server.py
third_party/sglang/python/sglang/srt/disaggregation/encode_receiver.py
third_party/sglang/python/sglang/srt/disaggregation/encode_server.py
third_party/sglang/python/sglang/srt/disaggregation/kv_events.py
third_party/sglang/python/sglang/srt/disaggregation/prefill.py
third_party/sglang/python/sglang/srt/disaggregation/utils.py
third_party/sglang/python/sglang/srt/disaggregation/ascend/__init__.py
third_party/sglang/python/sglang/srt/disaggregation/ascend/conn.py
third_party/sglang/python/sglang/srt/disaggregation/ascend/transfer_engine.py
third_party/sglang/python/sglang/srt/disaggregation/base/__init__.py
third_party/sglang/python/sglang/srt/disaggregation/base/conn.py
third_party/sglang/python/sglang/srt/disaggregation/common/__init__.py
third_party/sglang/python/sglang/srt/disaggregation/common/conn.py
third_party/sglang/python/sglang/srt/disaggregation/common/staging_buffer.py
third_party/sglang/python/sglang/srt/disaggregation/common/staging_handler.py
third_party/sglang/python/sglang/srt/disaggregation/common/utils.py
third_party/sglang/python/sglang/srt/disaggregation/fake/__init__.py
third_party/sglang/python/sglang/srt/disaggregation/fake/conn.py
third_party/sglang/python/sglang/srt/disaggregation/mooncake/__init__.py
third_party/sglang/python/sglang/srt/disaggregation/mooncake/conn.py
third_party/sglang/python/sglang/srt/disaggregation/mooncake/utils.py
third_party/sglang/python/sglang/srt/disaggregation/mori/__init__.py
third_party/sglang/python/sglang/srt/disaggregation/mori/conn.py
third_party/sglang/python/sglang/srt/disaggregation/nixl/__init__.py
third_party/sglang/python/sglang/srt/disaggregation/nixl/conn.py
third_party/sglang/python/sglang/srt/distributed/__init__.py
third_party/sglang/python/sglang/srt/distributed/communication_op.py
third_party/sglang/python/sglang/srt/distributed/naive_distributed.py
third_party/sglang/python/sglang/srt/distributed/parallel_state.py
third_party/sglang/python/sglang/srt/distributed/utils.py
third_party/sglang/python/sglang/srt/distributed/device_communicators/all_reduce_utils.py
third_party/sglang/python/sglang/srt/distributed/device_communicators/cuda_wrapper.py
third_party/sglang/python/sglang/srt/distributed/device_communicators/custom_all_reduce.py
third_party/sglang/python/sglang/srt/distributed/device_communicators/custom_all_reduce_ops.py
third_party/sglang/python/sglang/srt/distributed/device_communicators/custom_all_reduce_utils.py
third_party/sglang/python/sglang/srt/distributed/device_communicators/custom_all_reduce_v2.py
third_party/sglang/python/sglang/srt/distributed/device_communicators/hpu_communicator.py
third_party/sglang/python/sglang/srt/distributed/device_communicators/mooncake_transfer_engine.py
third_party/sglang/python/sglang/srt/distributed/device_communicators/npu_communicator.py
third_party/sglang/python/sglang/srt/distributed/device_communicators/pymscclpp.py
third_party/sglang/python/sglang/srt/distributed/device_communicators/pynccl.py
third_party/sglang/python/sglang/srt/distributed/device_communicators/pynccl_allocator.py
third_party/sglang/python/sglang/srt/distributed/device_communicators/pynccl_wrapper.py
third_party/sglang/python/sglang/srt/distributed/device_communicators/quick_all_reduce.py
third_party/sglang/python/sglang/srt/distributed/device_communicators/shm_broadcast.py
third_party/sglang/python/sglang/srt/distributed/device_communicators/torch_symm_mem.py
third_party/sglang/python/sglang/srt/distributed/device_communicators/xpu_communicator.py
third_party/sglang/python/sglang/srt/dllm/config.py
third_party/sglang/python/sglang/srt/dllm/algorithm/__init__.py
third_party/sglang/python/sglang/srt/dllm/algorithm/base.py
third_party/sglang/python/sglang/srt/dllm/algorithm/joint_threshold.py
third_party/sglang/python/sglang/srt/dllm/algorithm/low_confidence.py
third_party/sglang/python/sglang/srt/dllm/mixin/req.py
third_party/sglang/python/sglang/srt/dllm/mixin/scheduler.py
third_party/sglang/python/sglang/srt/elastic_ep/elastic_ep.py
third_party/sglang/python/sglang/srt/elastic_ep/expert_backup_client.py
third_party/sglang/python/sglang/srt/elastic_ep/expert_backup_manager.py
third_party/sglang/python/sglang/srt/entrypoints/EngineBase.py
third_party/sglang/python/sglang/srt/entrypoints/context.py
third_party/sglang/python/sglang/srt/entrypoints/engine.py
third_party/sglang/python/sglang/srt/entrypoints/engine_info_bootstrap_server.py
third_party/sglang/python/sglang/srt/entrypoints/engine_score_mixin.py
third_party/sglang/python/sglang/srt/entrypoints/grpc_server.py
third_party/sglang/python/sglang/srt/entrypoints/harmony_utils.py
third_party/sglang/python/sglang/srt/entrypoints/http_server.py
third_party/sglang/python/sglang/srt/entrypoints/http_server_engine.py
third_party/sglang/python/sglang/srt/entrypoints/ssl_utils.py
third_party/sglang/python/sglang/srt/entrypoints/tool.py
third_party/sglang/python/sglang/srt/entrypoints/v1_loads.py
third_party/sglang/python/sglang/srt/entrypoints/warmup.py
third_party/sglang/python/sglang/srt/entrypoints/anthropic/__init__.py
third_party/sglang/python/sglang/srt/entrypoints/anthropic/protocol.py
third_party/sglang/python/sglang/srt/entrypoints/anthropic/serving.py
third_party/sglang/python/sglang/srt/entrypoints/ollama/README.md
third_party/sglang/python/sglang/srt/entrypoints/ollama/__init__.py
third_party/sglang/python/sglang/srt/entrypoints/ollama/protocol.py
third_party/sglang/python/sglang/srt/entrypoints/ollama/serving.py
third_party/sglang/python/sglang/srt/entrypoints/ollama/smart_router.py
third_party/sglang/python/sglang/srt/entrypoints/openai/__init__.py
third_party/sglang/python/sglang/srt/entrypoints/openai/encoding_dsv32.py
third_party/sglang/python/sglang/srt/entrypoints/openai/protocol.py
third_party/sglang/python/sglang/srt/entrypoints/openai/serving_base.py
third_party/sglang/python/sglang/srt/entrypoints/openai/serving_chat.py
third_party/sglang/python/sglang/srt/entrypoints/openai/serving_classify.py
third_party/sglang/python/sglang/srt/entrypoints/openai/serving_completions.py
third_party/sglang/python/sglang/srt/entrypoints/openai/serving_embedding.py
third_party/sglang/python/sglang/srt/entrypoints/openai/serving_rerank.py
third_party/sglang/python/sglang/srt/entrypoints/openai/serving_responses.py
third_party/sglang/python/sglang/srt/entrypoints/openai/serving_score.py
third_party/sglang/python/sglang/srt/entrypoints/openai/serving_tokenize.py
third_party/sglang/python/sglang/srt/entrypoints/openai/serving_transcription.py
third_party/sglang/python/sglang/srt/entrypoints/openai/tool_server.py
third_party/sglang/python/sglang/srt/entrypoints/openai/usage_processor.py
third_party/sglang/python/sglang/srt/entrypoints/openai/utils.py
third_party/sglang/python/sglang/srt/eplb/__init__.py
third_party/sglang/python/sglang/srt/eplb/eplb_manager.py
third_party/sglang/python/sglang/srt/eplb/expert_distribution.py
third_party/sglang/python/sglang/srt/eplb/expert_location.py
third_party/sglang/python/sglang/srt/eplb/expert_location_dispatch.py
third_party/sglang/python/sglang/srt/eplb/expert_location_updater.py
third_party/sglang/python/sglang/srt/eplb/eplb_algorithms/__init__.py
third_party/sglang/python/sglang/srt/eplb/eplb_algorithms/deepseek.py
third_party/sglang/python/sglang/srt/eplb/eplb_algorithms/deepseek_vec.py
third_party/sglang/python/sglang/srt/eplb/eplb_algorithms/elasticity_aware.py
third_party/sglang/python/sglang/srt/eplb/eplb_simulator/__init__.py
third_party/sglang/python/sglang/srt/eplb/eplb_simulator/reader.py
third_party/sglang/python/sglang/srt/function_call/base_format_detector.py
third_party/sglang/python/sglang/srt/function_call/core_types.py
third_party/sglang/python/sglang/srt/function_call/deepseekv31_detector.py
third_party/sglang/python/sglang/srt/function_call/deepseekv32_detector.py
third_party/sglang/python/sglang/srt/function_call/deepseekv3_detector.py
third_party/sglang/python/sglang/srt/function_call/function_call_parser.py
third_party/sglang/python/sglang/srt/function_call/gigachat3_detector.py
third_party/sglang/python/sglang/srt/function_call/glm47_moe_detector.py
third_party/sglang/python/sglang/srt/function_call/glm4_moe_detector.py
third_party/sglang/python/sglang/srt/function_call/gpt_oss_detector.py
third_party/sglang/python/sglang/srt/function_call/hermes_detector.py
third_party/sglang/python/sglang/srt/function_call/internlm_detector.py
third_party/sglang/python/sglang/srt/function_call/json_array_parser.py
third_party/sglang/python/sglang/srt/function_call/kimik2_detector.py
third_party/sglang/python/sglang/srt/function_call/lfm2_detector.py
third_party/sglang/python/sglang/srt/function_call/llama32_detector.py
third_party/sglang/python/sglang/srt/function_call/mimo_detector.py
third_party/sglang/python/sglang/srt/function_call/minimax_m2.py
third_party/sglang/python/sglang/srt/function_call/mistral_detector.py
third_party/sglang/python/sglang/srt/function_call/pythonic_detector.py
third_party/sglang/python/sglang/srt/function_call/qwen25_detector.py
third_party/sglang/python/sglang/srt/function_call/qwen3_coder_detector.py
third_party/sglang/python/sglang/srt/function_call/step3_detector.py
third_party/sglang/python/sglang/srt/function_call/trinity_detector.py
third_party/sglang/python/sglang/srt/function_call/utils.py
third_party/sglang/python/sglang/srt/grpc/__init__.py
third_party/sglang/python/sglang/srt/hardware_backend/mlx/model_runner.py
third_party/sglang/python/sglang/srt/hardware_backend/mlx/model_runner_stub.py
third_party/sglang/python/sglang/srt/hardware_backend/mlx/tp_worker.py
third_party/sglang/python/sglang/srt/hardware_backend/npu/allocator_npu.py
third_party/sglang/python/sglang/srt/hardware_backend/npu/cmo.py
third_party/sglang/python/sglang/srt/hardware_backend/npu/memory_pool_npu.py
third_party/sglang/python/sglang/srt/hardware_backend/npu/utils.py
third_party/sglang/python/sglang/srt/hardware_backend/npu/attention/ascend_backend.py
third_party/sglang/python/sglang/srt/hardware_backend/npu/attention/ascend_torch_native_backend.py
third_party/sglang/python/sglang/srt/hardware_backend/npu/attention/mla_preprocess.py
third_party/sglang/python/sglang/srt/hardware_backend/npu/graph_runner/eagle_draft_extend_npu_graph_runner.py
third_party/sglang/python/sglang/srt/hardware_backend/npu/graph_runner/eagle_draft_npu_graph_runner.py
third_party/sglang/python/sglang/srt/hardware_backend/npu/graph_runner/npu_graph_runner.py
third_party/sglang/python/sglang/srt/hardware_backend/npu/graph_runner/vit_npu_graph_runner.py
third_party/sglang/python/sglang/srt/hardware_backend/npu/modules/deepseek_v2_attention_mla_npu.py
third_party/sglang/python/sglang/srt/hardware_backend/npu/modules/qwen_vl_processor.py
third_party/sglang/python/sglang/srt/hardware_backend/npu/moe/topk.py
third_party/sglang/python/sglang/srt/hardware_backend/npu/quantization/fused_moe_method_npu.py
third_party/sglang/python/sglang/srt/hardware_backend/npu/quantization/linear_method_npu.py
third_party/sglang/python/sglang/srt/layers/activation.py
third_party/sglang/python/sglang/srt/layers/amx_utils.py
third_party/sglang/python/sglang/srt/layers/communicator.py
third_party/sglang/python/sglang/srt/layers/communicator_nsa_cp.py
third_party/sglang/python/sglang/srt/layers/conv.py
third_party/sglang/python/sglang/srt/layers/dp_attention.py
third_party/sglang/python/sglang/srt/layers/elementwise.py
third_party/sglang/python/sglang/srt/layers/flashinfer_comm_fusion.py
third_party/sglang/python/sglang/srt/layers/int4fp8_utils.py
third_party/sglang/python/sglang/srt/layers/layernorm.py
third_party/sglang/python/sglang/srt/layers/linear.py
third_party/sglang/python/sglang/srt/layers/logits_processor.py
third_party/sglang/python/sglang/srt/layers/model_parallel.py
third_party/sglang/python/sglang/srt/layers/modelopt_utils.py
third_party/sglang/python/sglang/srt/layers/multimodal.py
third_party/sglang/python/sglang/srt/layers/n_gram_embedding.py
third_party/sglang/python/sglang/srt/layers/on_policy_utils.py
third_party/sglang/python/sglang/srt/layers/parameter.py
third_party/sglang/python/sglang/srt/layers/pooler.py
third_party/sglang/python/sglang/srt/layers/radix_attention.py
third_party/sglang/python/sglang/srt/layers/radix_linear_attention.py
third_party/sglang/python/sglang/srt/layers/rocm_linear_utils.py
third_party/sglang/python/sglang/srt/layers/sampler.py
third_party/sglang/python/sglang/srt/layers/sparse_pooler.py
third_party/sglang/python/sglang/srt/layers/torchao_utils.py
third_party/sglang/python/sglang/srt/layers/vocab_parallel_embedding.py
third_party/sglang/python/sglang/srt/layers/attention/aiter_backend.py
third_party/sglang/python/sglang/srt/layers/attention/attention_registry.py
third_party/sglang/python/sglang/srt/layers/attention/base_attn_backend.py
third_party/sglang/python/sglang/srt/layers/attention/cutlass_mla_backend.py
third_party/sglang/python/sglang/srt/layers/attention/double_sparsity_backend.py
third_party/sglang/python/sglang/srt/layers/attention/dual_chunk_flashattention_backend.py
third_party/sglang/python/sglang/srt/layers/attention/flashattention_backend.py
third_party/sglang/python/sglang/srt/layers/attention/flashinfer_backend.py
third_party/sglang/python/sglang/srt/layers/attention/flashinfer_mla_backend.py
third_party/sglang/python/sglang/srt/layers/attention/flashmla_backend.py
third_party/sglang/python/sglang/srt/layers/attention/hybrid_attn_backend.py
third_party/sglang/python/sglang/srt/layers/attention/hybrid_linear_attn_backend.py
third_party/sglang/python/sglang/srt/layers/attention/intel_amx_backend.py
third_party/sglang/python/sglang/srt/layers/attention/merge_state.py
third_party/sglang/python/sglang/srt/layers/attention/nsa_backend.py
third_party/sglang/python/sglang/srt/layers/attention/tbo_backend.py
third_party/sglang/python/sglang/srt/layers/attention/torch_flex_backend.py
third_party/sglang/python/sglang/srt/layers/attention/torch_native_backend.py
third_party/sglang/python/sglang/srt/layers/attention/triton_backend.py
third_party/sglang/python/sglang/srt/layers/attention/trtllm_mha_backend.py
third_party/sglang/python/sglang/srt/layers/attention/trtllm_mla_backend.py
third_party/sglang/python/sglang/srt/layers/attention/utils.py
third_party/sglang/python/sglang/srt/layers/attention/vision.py
third_party/sglang/python/sglang/srt/layers/attention/vision_utils.py
third_party/sglang/python/sglang/srt/layers/attention/wave_backend.py
third_party/sglang/python/sglang/srt/layers/attention/xpu_backend.py
third_party/sglang/python/sglang/srt/layers/attention/fla/chunk.py
third_party/sglang/python/sglang/srt/layers/attention/fla/chunk_delta_h.py
third_party/sglang/python/sglang/srt/layers/attention/fla/chunk_fwd.py
third_party/sglang/python/sglang/srt/layers/attention/fla/chunk_intra.py
third_party/sglang/python/sglang/srt/layers/attention/fla/chunk_intra_token_parallel.py
third_party/sglang/python/sglang/srt/layers/attention/fla/chunk_o.py
third_party/sglang/python/sglang/srt/layers/attention/fla/chunk_scaled_dot_kkt.py
third_party/sglang/python/sglang/srt/layers/attention/fla/cumsum.py
third_party/sglang/python/sglang/srt/layers/attention/fla/fused_gdn_gating.py
third_party/sglang/python/sglang/srt/layers/attention/fla/fused_norm_gate.py
third_party/sglang/python/sglang/srt/layers/attention/fla/fused_recurrent.py
third_party/sglang/python/sglang/srt/layers/attention/fla/fused_sigmoid_gating_recurrent.py
third_party/sglang/python/sglang/srt/layers/attention/fla/index.py
third_party/sglang/python/sglang/srt/layers/attention/fla/kda.py
third_party/sglang/python/sglang/srt/layers/attention/fla/l2norm.py
third_party/sglang/python/sglang/srt/layers/attention/fla/layernorm_gated.py
third_party/sglang/python/sglang/srt/layers/attention/fla/op.py
third_party/sglang/python/sglang/srt/layers/attention/fla/solve_tril.py
third_party/sglang/python/sglang/srt/layers/attention/fla/utils.py
third_party/sglang/python/sglang/srt/layers/attention/fla/wy_fast.py
third_party/sglang/python/sglang/srt/layers/attention/linear/__init__.py
third_party/sglang/python/sglang/srt/layers/attention/linear/gdn_backend.py
third_party/sglang/python/sglang/srt/layers/attention/linear/kda_backend.py
third_party/sglang/python/sglang/srt/layers/attention/linear/lightning_attn.py
third_party/sglang/python/sglang/srt/layers/attention/linear/lightning_backend.py
third_party/sglang/python/sglang/srt/layers/attention/linear/linear_metadata.py
third_party/sglang/python/sglang/srt/layers/attention/linear/seg_la.py
third_party/sglang/python/sglang/srt/layers/attention/linear/utils.py
third_party/sglang/python/sglang/srt/layers/attention/linear/kernels/__init__.py
third_party/sglang/python/sglang/srt/layers/attention/linear/kernels/gdn_cutedsl.py
third_party/sglang/python/sglang/srt/layers/attention/linear/kernels/gdn_flashinfer.py
third_party/sglang/python/sglang/srt/layers/attention/linear/kernels/gdn_triton.py
third_party/sglang/python/sglang/srt/layers/attention/linear/kernels/kda_cutedsl.py
third_party/sglang/python/sglang/srt/layers/attention/linear/kernels/kda_triton.py
third_party/sglang/python/sglang/srt/layers/attention/linear/kernels/kernel_backend.py
third_party/sglang/python/sglang/srt/layers/attention/mamba/causal_conv1d.py
third_party/sglang/python/sglang/srt/layers/attention/mamba/causal_conv1d_triton.py
third_party/sglang/python/sglang/srt/layers/attention/mamba/mamba.py
third_party/sglang/python/sglang/srt/layers/attention/mamba/mamba2_metadata.py
third_party/sglang/python/sglang/srt/layers/attention/mamba/mamba_state_scatter_triton.py
third_party/sglang/python/sglang/srt/layers/attention/mamba/mixer2_rms_norm_gated.py
third_party/sglang/python/sglang/srt/layers/attention/mamba/ops/__init__.py
third_party/sglang/python/sglang/srt/layers/attention/mamba/ops/layernorm_gated.py
third_party/sglang/python/sglang/srt/layers/attention/mamba/ops/mamba_ssm.py
third_party/sglang/python/sglang/srt/layers/attention/mamba/ops/ssd_bmm.py
third_party/sglang/python/sglang/srt/layers/attention/mamba/ops/ssd_chunk_scan.py
third_party/sglang/python/sglang/srt/layers/attention/mamba/ops/ssd_chunk_state.py
third_party/sglang/python/sglang/srt/layers/attention/mamba/ops/ssd_combined.py
third_party/sglang/python/sglang/srt/layers/attention/mamba/ops/ssd_state_passing.py
third_party/sglang/python/sglang/srt/layers/attention/mamba/ops/ssu_dispatch.py
third_party/sglang/python/sglang/srt/layers/attention/nsa/dequant_k_cache.py
third_party/sglang/python/sglang/srt/layers/attention/nsa/index_buf_accessor.py
third_party/sglang/python/sglang/srt/layers/attention/nsa/nsa_backend_mtp_precompute.py
third_party/sglang/python/sglang/srt/layers/attention/nsa/nsa_indexer.py
third_party/sglang/python/sglang/srt/layers/attention/nsa/nsa_mtp_verification.py
third_party/sglang/python/sglang/srt/layers/attention/nsa/quant_k_cache.py
third_party/sglang/python/sglang/srt/layers/attention/nsa/tilelang_kernel.py
third_party/sglang/python/sglang/srt/layers/attention/nsa/transform_index.py
third_party/sglang/python/sglang/srt/layers/attention/nsa/triton_kernel.py
third_party/sglang/python/sglang/srt/layers/attention/nsa/utils.py
third_party/sglang/python/sglang/srt/layers/attention/triton_ops/decode_attention.py
third_party/sglang/python/sglang/srt/layers/attention/triton_ops/double_sparsity_attention.py
third_party/sglang/python/sglang/srt/layers/attention/triton_ops/extend_attention.py
third_party/sglang/python/sglang/srt/layers/attention/triton_ops/merge_state.py
third_party/sglang/python/sglang/srt/layers/attention/triton_ops/prefill_attention.py
third_party/sglang/python/sglang/srt/layers/attention/triton_ops/rocm_mla_decode_rope.py
third_party/sglang/python/sglang/srt/layers/attention/triton_ops/trtllm_fp8_kv_kernel.py
third_party/sglang/python/sglang/srt/layers/attention/wave_ops/decode_attention.py
third_party/sglang/python/sglang/srt/layers/attention/wave_ops/extend_attention.py
third_party/sglang/python/sglang/srt/layers/attention/wave_ops/prefill_attention.py
third_party/sglang/python/sglang/srt/layers/deep_gemm_wrapper/__init__.py
third_party/sglang/python/sglang/srt/layers/deep_gemm_wrapper/compile_utils.py
third_party/sglang/python/sglang/srt/layers/deep_gemm_wrapper/configurer.py
third_party/sglang/python/sglang/srt/layers/deep_gemm_wrapper/entrypoint.py
third_party/sglang/python/sglang/srt/layers/moe/__init__.py
third_party/sglang/python/sglang/srt/layers/moe/cutlass_moe.py
third_party/sglang/python/sglang/srt/layers/moe/cutlass_moe_params.py
third_party/sglang/python/sglang/srt/layers/moe/cutlass_w4a8_moe.py
third_party/sglang/python/sglang/srt/layers/moe/flashinfer_cutedsl_moe.py
third_party/sglang/python/sglang/srt/layers/moe/flashinfer_trtllm_moe.py
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_native.py
third_party/sglang/python/sglang/srt/layers/moe/kt_ep_wrapper.py
third_party/sglang/python/sglang/srt/layers/moe/rocm_moe_utils.py
third_party/sglang/python/sglang/srt/layers/moe/routed_experts_capturer.py
third_party/sglang/python/sglang/srt/layers/moe/router.py
third_party/sglang/python/sglang/srt/layers/moe/topk.py
third_party/sglang/python/sglang/srt/layers/moe/utils.py
third_party/sglang/python/sglang/srt/layers/moe/ep_moe/__init__.py
third_party/sglang/python/sglang/srt/layers/moe/ep_moe/kernels.py
third_party/sglang/python/sglang/srt/layers/moe/ep_moe/layer.py
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/__init__.py
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/fused_marlin_moe.py
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/fused_moe.py
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/fused_moe_triton_config.py
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/fused_moe_triton_kernels.py
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/layer.py
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/moe_align_block_size.py
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/triton_kernels_moe.py
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/README.md
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=1,N=14336,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a16.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=1,N=14336,device_name=NVIDIA_A100-SXM4-80GB.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=1,N=1792,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a16.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=1,N=1792,device_name=NVIDIA_A100-SXM4-80GB.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=1,N=3072,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a16.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=1,N=3072,device_name=NVIDIA_H100_80GB_HBM3,dtype=int8_w8a16.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=1,N=3072,device_name=NVIDIA_H100_80GB_HBM3.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=1,N=3584,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a16.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=1,N=3584,device_name=NVIDIA_A100-SXM4-80GB.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=1,N=7168,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a16.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=1,N=7168,device_name=NVIDIA_A100-SXM4-80GB.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=144,N=512,device_name=NVIDIA_H100_80GB_HBM3.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=16,N=1024,device_name=NVIDIA_H100_80GB_HBM3.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=16,N=1024,device_name=NVIDIA_H200.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=16,N=1344,device_name=NVIDIA_A100-SXM4-40GB.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=16,N=1344,device_name=NVIDIA_A100-SXM4-80GB.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=16,N=1344,device_name=NVIDIA_H100_80GB_HBM3.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=16,N=14336,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a16.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=16,N=14336,device_name=NVIDIA_A100-SXM4-80GB.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=16,N=1792,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a16.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=16,N=1792,device_name=NVIDIA_A100-SXM4-80GB.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=16,N=2048,device_name=NVIDIA_H100_80GB_HBM3.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=16,N=2688,device_name=NVIDIA_A100-SXM4-80GB.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=16,N=2688,device_name=NVIDIA_H100_80GB_HBM3.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=16,N=3072,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a16.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=16,N=3072,device_name=NVIDIA_H100_80GB_HBM3,dtype=int8_w8a16.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=16,N=3200,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=16,N=3584,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a16.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=16,N=3584,device_name=NVIDIA_A100-SXM4-80GB.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=16,N=6400,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=16,N=7168,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a16.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=16,N=7168,device_name=NVIDIA_A100-SXM4-80GB.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=16,N=7168,device_name=NVIDIA_H100_80GB_HBM3,dtype=int8_w8a16.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=16,N=800,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=160,N=192,device_name=NVIDIA_A800-SXM4-80GB.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=20,N=2048,device_name=NVIDIA_H100_80GB_HBM3.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=24,N=1024,device_name=NVIDIA_H100_80GB_HBM3.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=256,N=128,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=256,N=128,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a8.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=256,N=128,device_name=NVIDIA_A800-SXM4-80GB,dtype=int8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=256,N=128,device_name=NVIDIA_A800-SXM4-80GB,dtype=int8_w8a8.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=256,N=128,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=256,N=128,device_name=NVIDIA_H20,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=256,N=128,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=256,N=128,device_name=NVIDIA_L20Y,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=256,N=256,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=256,N=256,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=256,N=256,device_name=AMD_Radeon_Graphics,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=256,N=256,device_name=NVIDIA_B200,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=256,N=256,device_name=NVIDIA_H20,dtype=int8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=256,N=256,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=256,N=64,device_name=NVIDIA_A800-SXM4-80GB.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=256,N=64,device_name=NVIDIA_L20,dtype=int8_w8a8.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=256,N=64,device_name=NVIDIA_L40S,dtype=int8_w8a8.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=64,N=1024,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=64,N=1280,device_name=NVIDIA_A100-SXM4-80GB.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=64,N=1280,device_name=NVIDIA_A800-SXM4-80GB.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=64,N=1280,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=64,N=1280,device_name=NVIDIA_H100_80GB_HBM3.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=64,N=1280,device_name=NVIDIA_H200,dtype=fp8_w8a8.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=64,N=1280,device_name=NVIDIA_H200.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=64,N=2560,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=64,N=2560,device_name=NVIDIA_H200,dtype=fp8_w8a8.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=64,N=2560,device_name=NVIDIA_H200.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=64,N=320,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=64,N=320,device_name=NVIDIA_H100_80GB_HBM3.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=64,N=320,device_name=NVIDIA_H200,dtype=fp8_w8a8.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=64,N=320,device_name=NVIDIA_H200.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=64,N=512,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=64,N=640,device_name=NVIDIA_A100-SXM4-80GB.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=64,N=640,device_name=NVIDIA_A800-SXM4-80GB.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=64,N=640,device_name=NVIDIA_GeForce_RTX_4090,dtype=fp8_w8a8.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=64,N=640,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=64,N=640,device_name=NVIDIA_H100_80GB_HBM3.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=64,N=640,device_name=NVIDIA_H200,dtype=fp8_w8a8.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=64,N=640,device_name=NVIDIA_H200.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=14336,device_name=AMD_Instinct_MI300X.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=14336,device_name=AMD_Instinct_MI325X.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=14336,device_name=AMD_Radeon_Graphics.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=14336,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=14336,device_name=NVIDIA_H200,dtype=fp8_w8a8.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=14336,device_name=NVIDIA_H200.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=1792,device_name=AMD_Instinct_MI300X.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=1792,device_name=AMD_Instinct_MI325X.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=1792,device_name=AMD_Radeon_Graphics.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=1792,device_name=NVIDIA_A100-SXM4-40GB.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=1792,device_name=NVIDIA_A100-SXM4-80GB.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=1792,device_name=NVIDIA_H100_80GB_HBM3.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=1792,device_name=NVIDIA_H200,dtype=fp8_w8a8.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=1792,device_name=NVIDIA_H200.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=2048,device_name=NVIDIA_A100-SXM4-80GB.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=2048,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=2048,device_name=NVIDIA_H100_80GB_HBM3.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=2048,device_name=NVIDIA_H200,dtype=fp8_w8a8.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=2048,device_name=NVIDIA_H200.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=3584,device_name=AMD_Instinct_MI300X.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=3584,device_name=AMD_Instinct_MI325X.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=3584,device_name=AMD_Radeon_Graphics.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=3584,device_name=NVIDIA_A100-SXM4-40GB.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=3584,device_name=NVIDIA_A100-SXM4-80GB.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=3584,device_name=NVIDIA_GeForce_RTX_4090,dtype=fp8_w8a8.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=3584,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=3584,device_name=NVIDIA_H100_80GB_HBM3.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=3584,device_name=NVIDIA_H200,dtype=fp8_w8a8.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=3584,device_name=NVIDIA_H200.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=3584,device_name=NVIDIA_L40S.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=4096,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=4096,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=4096,device_name=AMD_Radeon_Graphics,dtype=fp8_w8a8.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=4096,device_name=NVIDIA_A100-SXM4-80GB.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=4096,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=4096,device_name=NVIDIA_H100_80GB_HBM3.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=4096,device_name=NVIDIA_H200,dtype=fp8_w8a8.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=4096,device_name=NVIDIA_H200.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=7168,device_name=AMD_Instinct_MI300X.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=7168,device_name=AMD_Instinct_MI325X.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=7168,device_name=AMD_Radeon_Graphics.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=7168,device_name=NVIDIA_A100-SXM4-80GB.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=7168,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=7168,device_name=NVIDIA_H100_80GB_HBM3.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=7168,device_name=NVIDIA_H200,dtype=fp8_w8a8.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=7168,device_name=NVIDIA_H200.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=8192,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=8192,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=8192,device_name=AMD_Radeon_Graphics,dtype=fp8_w8a8.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=8192,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=8,N=8192,device_name=NVIDIA_H200,dtype=fp8_w8a8.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=128,N=192,device_name=NVIDIA_A800-SXM4-80GB.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=128,N=192,device_name=NVIDIA_H100_80GB_HBM3.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=128,N=192,device_name=NVIDIA_H20.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=128,N=192,device_name=NVIDIA_H200.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=128,N=384,device_name=NVIDIA_H100_80GB_HBM3.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=128,N=384,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=128,N=384,device_name=NVIDIA_H20.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=128,N=384,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=128,N=384,device_name=NVIDIA_H200.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=128,N=512,device_name=NVIDIA_H100_80GB_HBM3.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=128,N=768,device_name=NVIDIA_A800-SXM4-80GB.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=128,N=768,device_name=NVIDIA_H100_80GB_HBM3.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=128,N=768,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=128,N=768,device_name=NVIDIA_H20.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=128,N=768,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=128,N=768,device_name=NVIDIA_H200.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=128,N=96,device_name=NVIDIA_H20.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=129,N=352,device_name=NVIDIA_A800-SXM4-80GB,dtype=int8_w8a8.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=160,N=320,device_name=NVIDIA_A800-SXM4-80GB,dtype=int8_w8a8.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=161,N=192,device_name=NVIDIA_A800-SXM4-80GB,dtype=int8_w8a8.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=257,N=128,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=257,N=128,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=257,N=256,device_name=NVIDIA_B200,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=257,N=256,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=257,N=256,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=264,N=128,device_name=NVIDIA_A800-SXM4-80GB,dtype=int8_w8a8.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=264,N=256,device_name=NVIDIA_B200,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=264,N=256,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=264,N=256,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=272,N=128,device_name=NVIDIA_A800-SXM4-80GB,dtype=int8_w8a8.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=272,N=128,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=272,N=128,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=272,N=64,device_name=NVIDIA_A800-SXM4-80GB.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=288,N=64,device_name=NVIDIA_A800-SXM4-80GB.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=8,N=7168,device_name=NVIDIA_H100_80GB_HBM3.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_3_0/E=16,N=1024,device_name=NVIDIA_B200.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_3_1/E=128,N=352,device_name=NVIDIA_RTX_6000_Ada_Generation,dtype=fp8_w8a8.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_3_1/E=128,N=384,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_3_1/E=128,N=384,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_3_1/E=128,N=768,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_3_1/E=128,N=768,device_name=NVIDIA_H20.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_3_1/E=160,N=192,device_name=NVIDIA_H200,dtype=fp8_w8a8.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_3_1/E=160,N=320,device_name=NVIDIA_H20-3e.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_3_1/E=160,N=384,device_name=NVIDIA_H200,dtype=fp8_w8a8.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_3_1/E=160,N=640,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_3_1/E=257,N=128,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_3_1/E=257,N=128,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_3_1/E=257,N=256,device_name=NVIDIA_B200,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_3_1/E=257,N=256,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_3_1/E=257,N=256,device_name=NVIDIA_H20-3e,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_3_1/E=257,N=256,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_3_1/E=384,N=128,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_3_1/E=384,N=128,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_3_1/E=384,N=256,device_name=NVIDIA_H20-3e,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_3_1/E=385,N=128,device_name=NVIDIA_B200,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_3_1/E=385,N=128,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_3_1/E=8,N=7168,device_name=NVIDIA_H100_80GB_HBM3.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=128,N=1856,device_name=NVIDIA_H100_80GB_HBM3.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=128,N=1856,device_name=NVIDIA_L40S.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=128,N=192,device_name=NVIDIA_H200,dtype=fp8_w8a8.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=128,N=352,device_name=NVIDIA_RTX_5880_Ada_Generation,dtype=fp8_w8a8.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=128,N=384,device_name=NVIDIA_B200,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=128,N=928,device_name=NVIDIA_L40S.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=129,N=352,device_name=NVIDIA_B200,dtype=fp8_w8a8.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=129,N=352,device_name=NVIDIA_RTX_PRO_6000_Blackwell_Max-Q_Workstation_Edition,dtype=fp8_w8a8.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=129,N=704,device_name=NVIDIA_B200,dtype=fp8_w8a8.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=160,N=384,device_name=NVIDIA_B200,dtype=fp8_w8a8.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=161,N=192,device_name=NVIDIA_H200,dtype=fp8_w8a8,per_channel_quant=True.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=161,N=384,device_name=NVIDIA_RTX_PRO_6000_Blackwell_Max-Q_Workstation_Edition,dtype=fp8_w8a8.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=256,N=256,device_name=NVIDIA_B200,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=256,N=256,device_name=NVIDIA_B200.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=256,N=256,device_name=NVIDIA_H800,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=256,N=512,device_name=NVIDIA_B200.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=256,N=512,device_name=NVIDIA_H20.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=257,N=128,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=257,N=256,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=257,N=256,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128]_down.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=257,N=64,device_name=NVIDIA_A100-SXM4-80GB.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=384,N=128,device_name=,dtype=int4_w4a16.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=384,N=128,device_name=,dtype=int4_w4a16_down.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=384,N=256,device_name=NVIDIA_B200,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=512,N=128,device_name=NVIDIA_H100_80GB_HBM3.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=512,N=128,device_name=NVIDIA_H20-3e.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=512,N=128,device_name=NVIDIA_H200.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=512,N=128,device_name=NVIDIA_H800,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=512,N=256,device_name=NVIDIA_B200.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=512,N=256,device_name=NVIDIA_H20-3e.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=512,N=256,device_name=NVIDIA_H200.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=512,N=64,device_name=NVIDIA_H100_80GB_HBM3.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=512,N=64,device_name=NVIDIA_H200.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=128,N=1344,device_name=NVIDIA_B200.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=128,N=1344,device_name=NVIDIA_H100_80GB_HBM3.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=128,N=1856,device_name=NVIDIA_B200.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=128,N=1856,device_name=NVIDIA_H100_80GB_HBM3.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=128,N=232,device_name=NVIDIA_B200.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=128,N=232,device_name=NVIDIA_H100_80GB_HBM3.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=128,N=2688,device_name=NVIDIA_B200.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=128,N=2688,device_name=NVIDIA_H100_80GB_HBM3.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=128,N=464,device_name=NVIDIA_B200.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=128,N=464,device_name=NVIDIA_H100_80GB_HBM3.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=128,N=928,device_name=NVIDIA_B200.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=128,N=928,device_name=NVIDIA_H100_80GB_HBM3.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=16,N=1856,device_name=NVIDIA_B200.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=16,N=1856,device_name=NVIDIA_H100_80GB_HBM3.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=16,N=2048,device_name=NVIDIA_B200.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=161,N=192,device_name=NVIDIA_B200,dtype=fp8_w8a8,per_channel_quant=True.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=161,N=192,device_name=NVIDIA_H20,dtype=fp8_w8a8,per_channel_quant=True.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=161,N=192,device_name=NVIDIA_H20-3e,dtype=fp8_w8a8,per_channel_quant=True.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=161,N=192,device_name=NVIDIA_H200,dtype=fp8_w8a8,per_channel_quant=True.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=161,N=192,device_name=NVIDIA_H200.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=161,N=192,device_name=NVIDIA_RTX_PRO_6000_Blackwell_Max-Q_Workstation_Edition,dtype=fp8_w8a8,per_channel_quant=True.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=161,N=192,device_name=NVIDIA_RTX_PRO_6000_Blackwell_Max-Q_Workstation_Edition.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=161,N=384,device_name=NVIDIA_B200,dtype=fp8_w8a8,per_channel_quant=True.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=161,N=384,device_name=NVIDIA_H200,dtype=fp8_w8a8,per_channel_quant=True.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=20,N=1536,device_name=NVIDIA_H200,dtype=fp8_w8a8,per_channel_quant=True.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=20,N=1536,device_name=NVIDIA_H200.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=256,N=1344,device_name=NVIDIA_B200.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=256,N=1344,device_name=NVIDIA_H100_80GB_HBM3.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=256,N=2688,device_name=NVIDIA_B200.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=256,N=2688,device_name=NVIDIA_H100_80GB_HBM3.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=256,N=384,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=256,N=384,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=256,N=672,device_name=NVIDIA_B200.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=256,N=672,device_name=NVIDIA_H100_80GB_HBM3.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=257,N=256,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=257,N=256,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128]_down.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=257,N=256,device_name=NVIDIA_H20-3e,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=257,N=256,device_name=NVIDIA_H20-3e,dtype=fp8_w8a8,block_shape=[128, 128]_down.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=257,N=256,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=257,N=256,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128]_down.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=32,N=1856,device_name=NVIDIA_B200.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=32,N=1856,device_name=NVIDIA_H100_80GB_HBM3.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=32,N=928,device_name=NVIDIA_B200.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=32,N=928,device_name=NVIDIA_H100_80GB_HBM3.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=40,N=1536,device_name=NVIDIA_H200,dtype=fp8_w8a8,per_channel_quant=True.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=512,N=128,device_name=NVIDIA_B200.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=512,N=128,device_name=NVIDIA_H200.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=512,N=1344,device_name=NVIDIA_B200.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=512,N=1344,device_name=NVIDIA_H100_80GB_HBM3.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=512,N=256,device_name=NVIDIA_B200.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=512,N=256,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=512,N=256,device_name=NVIDIA_H200.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=512,N=2688,device_name=NVIDIA_B200.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=512,N=2688,device_name=NVIDIA_H100_80GB_HBM3.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=512,N=336,device_name=NVIDIA_B200.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=512,N=336,device_name=NVIDIA_H100_80GB_HBM3.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=512,N=672,device_name=NVIDIA_B200.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=512,N=672,device_name=NVIDIA_H100_80GB_HBM3.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=64,N=1856,device_name=NVIDIA_B200.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=64,N=1856,device_name=NVIDIA_H100_80GB_HBM3.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=64,N=2688,device_name=NVIDIA_B200.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=64,N=2688,device_name=NVIDIA_H100_80GB_HBM3.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=64,N=464,device_name=NVIDIA_B200.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=64,N=464,device_name=NVIDIA_H100_80GB_HBM3.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=64,N=928,device_name=NVIDIA_B200.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=64,N=928,device_name=NVIDIA_H100_80GB_HBM3.json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=80,N=640,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=80,N=640,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128]_down.json
third_party/sglang/python/sglang/srt/layers/moe/moe_runner/__init__.py
third_party/sglang/python/sglang/srt/layers/moe/moe_runner/base.py
third_party/sglang/python/sglang/srt/layers/moe/moe_runner/deep_gemm.py
third_party/sglang/python/sglang/srt/layers/moe/moe_runner/flashinfer_trtllm.py
third_party/sglang/python/sglang/srt/layers/moe/moe_runner/marlin.py
third_party/sglang/python/sglang/srt/layers/moe/moe_runner/runner.py
third_party/sglang/python/sglang/srt/layers/moe/moe_runner/triton.py
third_party/sglang/python/sglang/srt/layers/moe/moe_runner/triton_kernels.py
third_party/sglang/python/sglang/srt/layers/moe/token_dispatcher/__init__.py
third_party/sglang/python/sglang/srt/layers/moe/token_dispatcher/base.py
third_party/sglang/python/sglang/srt/layers/moe/token_dispatcher/deepep.py
third_party/sglang/python/sglang/srt/layers/moe/token_dispatcher/flashinfer.py
third_party/sglang/python/sglang/srt/layers/moe/token_dispatcher/flashinfer_utils.py
third_party/sglang/python/sglang/srt/layers/moe/token_dispatcher/fuseep.py
third_party/sglang/python/sglang/srt/layers/moe/token_dispatcher/mooncake.py
third_party/sglang/python/sglang/srt/layers/moe/token_dispatcher/moriep.py
third_party/sglang/python/sglang/srt/layers/moe/token_dispatcher/nixl.py
third_party/sglang/python/sglang/srt/layers/moe/token_dispatcher/standard.py
third_party/sglang/python/sglang/srt/layers/quantization/__init__.py
third_party/sglang/python/sglang/srt/layers/quantization/auto_round.py
third_party/sglang/python/sglang/srt/layers/quantization/awq.py
third_party/sglang/python/sglang/srt/layers/quantization/awq_triton.py
third_party/sglang/python/sglang/srt/layers/quantization/base_config.py
third_party/sglang/python/sglang/srt/layers/quantization/base_scheme.py
third_party/sglang/python/sglang/srt/layers/quantization/bitsandbytes.py
third_party/sglang/python/sglang/srt/layers/quantization/blockwise_int8.py
third_party/sglang/python/sglang/srt/layers/quantization/fp4_utils.py
third_party/sglang/python/sglang/srt/layers/quantization/fp8.py
third_party/sglang/python/sglang/srt/layers/quantization/fp8_kernel.py
third_party/sglang/python/sglang/srt/layers/quantization/fp8_utils.py
third_party/sglang/python/sglang/srt/layers/quantization/fpgemm_fp8.py
third_party/sglang/python/sglang/srt/layers/quantization/gguf.py
third_party/sglang/python/sglang/srt/layers/quantization/gptq.py
third_party/sglang/python/sglang/srt/layers/quantization/int8_kernel.py
third_party/sglang/python/sglang/srt/layers/quantization/int8_utils.py
third_party/sglang/python/sglang/srt/layers/quantization/kv_cache.py
third_party/sglang/python/sglang/srt/layers/quantization/kvfp4_tensor.py
third_party/sglang/python/sglang/srt/layers/quantization/marlin_utils.py
third_party/sglang/python/sglang/srt/layers/quantization/marlin_utils_fp8.py
third_party/sglang/python/sglang/srt/layers/quantization/modelopt_quant.py
third_party/sglang/python/sglang/srt/layers/quantization/moe_wna16.py
third_party/sglang/python/sglang/srt/layers/quantization/mxfp4.py
third_party/sglang/python/sglang/srt/layers/quantization/mxfp4_tensor.py
third_party/sglang/python/sglang/srt/layers/quantization/petit.py
third_party/sglang/python/sglang/srt/layers/quantization/petit_utils.py
third_party/sglang/python/sglang/srt/layers/quantization/qoq.py
third_party/sglang/python/sglang/srt/layers/quantization/quark_int4fp8_moe.py
third_party/sglang/python/sglang/srt/layers/quantization/rocm_mxfp4_utils.py
third_party/sglang/python/sglang/srt/layers/quantization/unquant.py
third_party/sglang/python/sglang/srt/layers/quantization/utils.py
third_party/sglang/python/sglang/srt/layers/quantization/w4afp8.py
third_party/sglang/python/sglang/srt/layers/quantization/w8a8_fp8.py
third_party/sglang/python/sglang/srt/layers/quantization/w8a8_int8.py
third_party/sglang/python/sglang/srt/layers/quantization/compressed_tensors/README.md
third_party/sglang/python/sglang/srt/layers/quantization/compressed_tensors/compressed_tensors.py
third_party/sglang/python/sglang/srt/layers/quantization/compressed_tensors/utils.py
third_party/sglang/python/sglang/srt/layers/quantization/compressed_tensors/schemes/__init__.py
third_party/sglang/python/sglang/srt/layers/quantization/compressed_tensors/schemes/compressed_tensors_scheme.py
third_party/sglang/python/sglang/srt/layers/quantization/compressed_tensors/schemes/compressed_tensors_w4a4_mxint4_moe.py
third_party/sglang/python/sglang/srt/layers/quantization/compressed_tensors/schemes/compressed_tensors_w4a4_nvfp4.py
third_party/sglang/python/sglang/srt/layers/quantization/compressed_tensors/schemes/compressed_tensors_w4a4_nvfp4_moe.py
third_party/sglang/python/sglang/srt/layers/quantization/compressed_tensors/schemes/compressed_tensors_w4a8_int8_moe.py
third_party/sglang/python/sglang/srt/layers/quantization/compressed_tensors/schemes/compressed_tensors_w8a16_fp8.py
third_party/sglang/python/sglang/srt/layers/quantization/compressed_tensors/schemes/compressed_tensors_w8a8_fp8.py
third_party/sglang/python/sglang/srt/layers/quantization/compressed_tensors/schemes/compressed_tensors_w8a8_fp8_moe.py
third_party/sglang/python/sglang/srt/layers/quantization/compressed_tensors/schemes/compressed_tensors_w8a8_int8.py
third_party/sglang/python/sglang/srt/layers/quantization/compressed_tensors/schemes/compressed_tensors_w8a8_int8_moe.py
third_party/sglang/python/sglang/srt/layers/quantization/compressed_tensors/schemes/compressed_tensors_wNa16.py
third_party/sglang/python/sglang/srt/layers/quantization/compressed_tensors/schemes/compressed_tensors_wNa16_moe.py
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=1280,K=5120,device_name=NVIDIA_B200,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=1536,K=1536,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=1536,K=1536,device_name=NVIDIA_A800-SXM4-80GB,dtype=int8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=1536,K=1536,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=1536,K=1536,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=1536,K=1536,device_name=NVIDIA_L20Y,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=1536,K=7168,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=1536,K=7168,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=1536,K=7168,device_name=AMD_Radeon_Graphics,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=1536,K=7168,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=1536,K=7168,device_name=NVIDIA_A800-SXM4-80GB,dtype=int8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=1536,K=7168,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=1536,K=7168,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=1536,K=7168,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=1536,K=7168,device_name=NVIDIA_L20Y,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=2048,K=4096,device_name=NVIDIA_L40,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=2048,K=512,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=2048,K=512,device_name=NVIDIA_A800-SXM4-80GB,dtype=int8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=2048,K=512,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=2048,K=512,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=2048,K=512,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=2048,K=512,device_name=NVIDIA_L20Y,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=2304,K=7168,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=2304,K=7168,device_name=NVIDIA_A800-SXM4-80GB,dtype=int8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=2304,K=7168,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=2304,K=7168,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=2304,K=7168,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=2304,K=7168,device_name=NVIDIA_L20Y,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=24576,K=1536,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=24576,K=1536,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=24576,K=1536,device_name=AMD_Radeon_Graphics,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=24576,K=7168,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=24576,K=7168,device_name=NVIDIA_A800-SXM4-80GB,dtype=int8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=24576,K=7168,device_name=NVIDIA_B200,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=24576,K=7168,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=24576,K=7168,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=24576,K=7168,device_name=NVIDIA_H20,dtype=int8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=24576,K=7168,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=24576,K=7168,device_name=NVIDIA_L20Y,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=256,K=7168,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=256,K=7168,device_name=NVIDIA_A800-SXM4-80GB,dtype=int8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=256,K=7168,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=256,K=7168,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=256,K=7168,device_name=NVIDIA_L20Y,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=3072,K=1536,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=3072,K=1536,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=3072,K=1536,device_name=AMD_Radeon_Graphics,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=3072,K=1536,device_name=NVIDIA_B200,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=3072,K=1536,device_name=NVIDIA_H20,dtype=int8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=3072,K=1536,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=3072,K=7168,device_name=NVIDIA_B200,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=3072,K=7168,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=3072,K=7168,device_name=NVIDIA_H20,dtype=int8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=3072,K=7168,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=32768,K=512,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=32768,K=512,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=32768,K=512,device_name=AMD_Radeon_Graphics,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=32768,K=512,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=32768,K=512,device_name=NVIDIA_A800-SXM4-80GB,dtype=int8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=32768,K=512,device_name=NVIDIA_B200,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=32768,K=512,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=32768,K=512,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=32768,K=512,device_name=NVIDIA_H20,dtype=int8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=32768,K=512,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=32768,K=512,device_name=NVIDIA_L20Y,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=36864,K=7168,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=36864,K=7168,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=4096,K=512,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=4096,K=512,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=4096,K=512,device_name=AMD_Radeon_Graphics,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=4096,K=512,device_name=NVIDIA_B200,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=4096,K=512,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=4096,K=512,device_name=NVIDIA_H20,dtype=int8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=4096,K=512,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=4608,K=7168,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=4608,K=7168,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=4608,K=7168,device_name=AMD_Radeon_Graphics,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=4608,K=7168,device_name=NVIDIA_B200,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=4608,K=7168,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=4608,K=7168,device_name=NVIDIA_H20,dtype=int8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=4608,K=7168,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=512,K=7168,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=512,K=7168,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=512,K=7168,device_name=AMD_Radeon_Graphics,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=512,K=7168,device_name=NVIDIA_B200,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=512,K=7168,device_name=NVIDIA_H20,dtype=int8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=512,K=7168,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=5120,K=1024,device_name=NVIDIA_B200,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=5120,K=2048,device_name=NVIDIA_L40,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=5120,K=3200,device_name=NVIDIA_B200,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=576,K=7168,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=576,K=7168,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=576,K=7168,device_name=AMD_Radeon_Graphics,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=576,K=7168,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=576,K=7168,device_name=NVIDIA_A800-SXM4-80GB,dtype=int8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=576,K=7168,device_name=NVIDIA_B200,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=576,K=7168,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=576,K=7168,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=576,K=7168,device_name=NVIDIA_H20,dtype=int8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=576,K=7168,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=576,K=7168,device_name=NVIDIA_L20Y,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=6400,K=5120,device_name=NVIDIA_B200,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=7168,K=1024,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=7168,K=1024,device_name=NVIDIA_A800-SXM4-80GB,dtype=int8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=7168,K=1024,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=7168,K=1024,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=7168,K=1024,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=7168,K=1024,device_name=NVIDIA_L20Y,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=7168,K=1152,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=7168,K=1152,device_name=NVIDIA_A800-SXM4-80GB,dtype=int8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=7168,K=1152,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=7168,K=1152,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=7168,K=1152,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=7168,K=1152,device_name=NVIDIA_L20Y,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=7168,K=128,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=7168,K=128,device_name=NVIDIA_A800-SXM4-80GB,dtype=int8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=7168,K=128,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=7168,K=128,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=7168,K=128,device_name=NVIDIA_L20Y,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=7168,K=16384,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=7168,K=16384,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=7168,K=16384,device_name=AMD_Radeon_Graphics,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=7168,K=16384,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=7168,K=16384,device_name=NVIDIA_A800-SXM4-80GB,dtype=int8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=7168,K=16384,device_name=NVIDIA_B200,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=7168,K=16384,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=7168,K=16384,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=7168,K=16384,device_name=NVIDIA_H20,dtype=int8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=7168,K=16384,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=7168,K=16384,device_name=NVIDIA_L20Y,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=7168,K=18432,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=7168,K=18432,device_name=NVIDIA_A800-SXM4-80GB,dtype=int8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=7168,K=18432,device_name=NVIDIA_B200,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=7168,K=18432,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=7168,K=18432,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=7168,K=18432,device_name=NVIDIA_H20,dtype=int8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=7168,K=18432,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=7168,K=18432,device_name=NVIDIA_L20Y,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=7168,K=2048,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=7168,K=2048,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=7168,K=2048,device_name=AMD_Radeon_Graphics,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=7168,K=2048,device_name=NVIDIA_B200,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=7168,K=2048,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=7168,K=2048,device_name=NVIDIA_H20,dtype=int8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=7168,K=2048,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=7168,K=2304,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=7168,K=2304,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=7168,K=2304,device_name=AMD_Radeon_Graphics,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=7168,K=2304,device_name=NVIDIA_B200,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=7168,K=2304,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=7168,K=2304,device_name=NVIDIA_H20,dtype=int8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=7168,K=2304,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=7168,K=256,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=7168,K=256,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=7168,K=256,device_name=AMD_Radeon_Graphics,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=7168,K=256,device_name=NVIDIA_B200,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=7168,K=256,device_name=NVIDIA_H20,dtype=int8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/N=7168,K=256,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json
third_party/sglang/python/sglang/srt/layers/quantization/configs/README.md
third_party/sglang/python/sglang/srt/layers/quantization/modelslim/README.md
third_party/sglang/python/sglang/srt/layers/quantization/modelslim/modelslim.py
third_party/sglang/python/sglang/srt/layers/quantization/modelslim/schemes/__init__.py
third_party/sglang/python/sglang/srt/layers/quantization/modelslim/schemes/modelslim_scheme.py
third_party/sglang/python/sglang/srt/layers/quantization/modelslim/schemes/modelslim_w4a4_int4.py
third_party/sglang/python/sglang/srt/layers/quantization/modelslim/schemes/modelslim_w4a4_int4_moe.py
third_party/sglang/python/sglang/srt/layers/quantization/modelslim/schemes/modelslim_w4a8_int8_moe.py
third_party/sglang/python/sglang/srt/layers/quantization/modelslim/schemes/modelslim_w8a8_int8.py
third_party/sglang/python/sglang/srt/layers/quantization/modelslim/schemes/modelslim_w8a8_int8_moe.py
third_party/sglang/python/sglang/srt/layers/quantization/quark/__init__.py
third_party/sglang/python/sglang/srt/layers/quantization/quark/quark.py
third_party/sglang/python/sglang/srt/layers/quantization/quark/utils.py
third_party/sglang/python/sglang/srt/layers/quantization/quark/schemes/__init__.py
third_party/sglang/python/sglang/srt/layers/quantization/quark/schemes/quark_scheme.py
third_party/sglang/python/sglang/srt/layers/quantization/quark/schemes/quark_w4a4_mxfp4.py
third_party/sglang/python/sglang/srt/layers/quantization/quark/schemes/quark_w4a4_mxfp4_moe.py
third_party/sglang/python/sglang/srt/layers/quantization/quark/schemes/quark_w8a8_fp8.py
third_party/sglang/python/sglang/srt/layers/quantization/quark/schemes/quark_w8a8_fp8_moe.py
third_party/sglang/python/sglang/srt/layers/rotary_embedding/__init__.py
third_party/sglang/python/sglang/srt/layers/rotary_embedding/base.py
third_party/sglang/python/sglang/srt/layers/rotary_embedding/factory.py
third_party/sglang/python/sglang/srt/layers/rotary_embedding/mrope.py
third_party/sglang/python/sglang/srt/layers/rotary_embedding/mrope_rope_index.py
third_party/sglang/python/sglang/srt/layers/rotary_embedding/rope_variant.py
third_party/sglang/python/sglang/srt/layers/rotary_embedding/triton_kernels.py
third_party/sglang/python/sglang/srt/layers/rotary_embedding/utils.py
third_party/sglang/python/sglang/srt/layers/rotary_embedding/yarn.py
third_party/sglang/python/sglang/srt/layers/utils/__init__.py
third_party/sglang/python/sglang/srt/layers/utils/common.py
third_party/sglang/python/sglang/srt/layers/utils/cp_utils.py
third_party/sglang/python/sglang/srt/layers/utils/hash.py
third_party/sglang/python/sglang/srt/layers/utils/logprob.py
third_party/sglang/python/sglang/srt/layers/utils/multi_platform.py
third_party/sglang/python/sglang/srt/lora/eviction_policy.py
third_party/sglang/python/sglang/srt/lora/layers.py
third_party/sglang/python/sglang/srt/lora/lora.py
third_party/sglang/python/sglang/srt/lora/lora_config.py
third_party/sglang/python/sglang/srt/lora/lora_manager.py
third_party/sglang/python/sglang/srt/lora/lora_moe_runner_marlin.py
third_party/sglang/python/sglang/srt/lora/lora_moe_runners.py
third_party/sglang/python/sglang/srt/lora/lora_overlap_loader.py
third_party/sglang/python/sglang/srt/lora/lora_registry.py
third_party/sglang/python/sglang/srt/lora/mem_pool.py
third_party/sglang/python/sglang/srt/lora/utils.py
third_party/sglang/python/sglang/srt/lora/backend/ascend_backend.py
third_party/sglang/python/sglang/srt/lora/backend/base_backend.py
third_party/sglang/python/sglang/srt/lora/backend/chunked_backend.py
third_party/sglang/python/sglang/srt/lora/backend/lmhead_mixing.py
third_party/sglang/python/sglang/srt/lora/backend/lora_registry.py
third_party/sglang/python/sglang/srt/lora/backend/torch_backend.py
third_party/sglang/python/sglang/srt/lora/backend/triton_backend.py
third_party/sglang/python/sglang/srt/lora/torch_ops/__init__.py
third_party/sglang/python/sglang/srt/lora/torch_ops/lora_ops.py
third_party/sglang/python/sglang/srt/lora/triton_ops/__init__.py
third_party/sglang/python/sglang/srt/lora/triton_ops/chunked_embedding_lora_a.py
third_party/sglang/python/sglang/srt/lora/triton_ops/chunked_sgmv_expand.py
third_party/sglang/python/sglang/srt/lora/triton_ops/chunked_sgmv_shrink.py
third_party/sglang/python/sglang/srt/lora/triton_ops/embedding_lora_a.py
third_party/sglang/python/sglang/srt/lora/triton_ops/fused_moe_lora_kernel.py
third_party/sglang/python/sglang/srt/lora/triton_ops/gate_up_lora_b.py
third_party/sglang/python/sglang/srt/lora/triton_ops/kernel_utils.py
third_party/sglang/python/sglang/srt/lora/triton_ops/lora_tuning_config.py
third_party/sglang/python/sglang/srt/lora/triton_ops/qkv_lora_b.py
third_party/sglang/python/sglang/srt/lora/triton_ops/sgemm_lora_a.py
third_party/sglang/python/sglang/srt/lora/triton_ops/sgemm_lora_b.py
third_party/sglang/python/sglang/srt/lora/triton_ops/virtual_experts.py
third_party/sglang/python/sglang/srt/lora/triton_ops/csgmv_configs/triton_3_5_1/lora_expand,K=1024,R=64,S=1,device=NVIDIA_H200.json
third_party/sglang/python/sglang/srt/lora/triton_ops/csgmv_configs/triton_3_5_1/lora_expand,K=4096,R=64,S=3,device=NVIDIA_H200.json
third_party/sglang/python/sglang/srt/lora/triton_ops/csgmv_configs/triton_3_5_1/lora_expand,K=6144,R=64,S=2,device=NVIDIA_H200.json
third_party/sglang/python/sglang/srt/lora/triton_ops/csgmv_configs/triton_3_5_1/lora_shrink,K=1024,R=64,S=2,device=NVIDIA_H200.json
third_party/sglang/python/sglang/srt/lora/triton_ops/csgmv_configs/triton_3_5_1/lora_shrink,K=1024,R=64,S=3,device=NVIDIA_H200.json
third_party/sglang/python/sglang/srt/lora/triton_ops/csgmv_configs/triton_3_5_1/lora_shrink,K=2048,R=64,S=1,device=NVIDIA_H200.json
third_party/sglang/python/sglang/srt/lora/triton_ops/csgmv_configs/triton_3_5_1/lora_shrink,K=3072,R=64,S=1,device=NVIDIA_H200.json
third_party/sglang/python/sglang/srt/managers/async_dynamic_batch_tokenizer.py
third_party/sglang/python/sglang/srt/managers/cache_controller.py
third_party/sglang/python/sglang/srt/managers/configure_logging.py
third_party/sglang/python/sglang/srt/managers/data_parallel_controller.py
third_party/sglang/python/sglang/srt/managers/detokenizer_manager.py
third_party/sglang/python/sglang/srt/managers/disagg_service.py
third_party/sglang/python/sglang/srt/managers/hisparse_coordinator.py
third_party/sglang/python/sglang/srt/managers/io_struct.py
third_party/sglang/python/sglang/srt/managers/mm_utils.py
third_party/sglang/python/sglang/srt/managers/multi_tokenizer_mixin.py
third_party/sglang/python/sglang/srt/managers/multimodal_processor.py
third_party/sglang/python/sglang/srt/managers/overlap_utils.py
third_party/sglang/python/sglang/srt/managers/prefill_delayer.py
third_party/sglang/python/sglang/srt/managers/schedule_batch.py
third_party/sglang/python/sglang/srt/managers/schedule_policy.py
third_party/sglang/python/sglang/srt/managers/scheduler.py
third_party/sglang/python/sglang/srt/managers/scheduler_dp_attn_mixin.py
third_party/sglang/python/sglang/srt/managers/scheduler_input_blocker.py
third_party/sglang/python/sglang/srt/managers/scheduler_output_processor_mixin.py
third_party/sglang/python/sglang/srt/managers/scheduler_pp_mixin.py
third_party/sglang/python/sglang/srt/managers/scheduler_profiler_mixin.py
third_party/sglang/python/sglang/srt/managers/scheduler_recv_skipper.py
third_party/sglang/python/sglang/srt/managers/scheduler_runtime_checker_mixin.py
third_party/sglang/python/sglang/srt/managers/scheduler_update_weights_mixin.py
third_party/sglang/python/sglang/srt/managers/session_controller.py
third_party/sglang/python/sglang/srt/managers/template_manager.py
third_party/sglang/python/sglang/srt/managers/tokenizer_communicator_mixin.py
third_party/sglang/python/sglang/srt/managers/tokenizer_manager.py
third_party/sglang/python/sglang/srt/managers/tokenizer_manager_score_mixin.py
third_party/sglang/python/sglang/srt/managers/tp_worker.py
third_party/sglang/python/sglang/srt/managers/utils.py
third_party/sglang/python/sglang/srt/mem_cache/allocator.py
third_party/sglang/python/sglang/srt/mem_cache/base_prefix_cache.py
third_party/sglang/python/sglang/srt/mem_cache/cache_init_params.py
third_party/sglang/python/sglang/srt/mem_cache/chunk_cache.py
third_party/sglang/python/sglang/srt/mem_cache/common.py
third_party/sglang/python/sglang/srt/mem_cache/evict_policy.py
third_party/sglang/python/sglang/srt/mem_cache/flush_cache.py
third_party/sglang/python/sglang/srt/mem_cache/hi_mamba_radix_cache.py
third_party/sglang/python/sglang/srt/mem_cache/hicache_storage.py
third_party/sglang/python/sglang/srt/mem_cache/hiradix_cache.py
third_party/sglang/python/sglang/srt/mem_cache/hisparse_memory_pool.py
third_party/sglang/python/sglang/srt/mem_cache/mamba_radix_cache.py
third_party/sglang/python/sglang/srt/mem_cache/memory_pool.py
third_party/sglang/python/sglang/srt/mem_cache/memory_pool_host.py
third_party/sglang/python/sglang/srt/mem_cache/multimodal_cache.py
third_party/sglang/python/sglang/srt/mem_cache/radix_cache.py
third_party/sglang/python/sglang/srt/mem_cache/radix_cache_cpp.py
third_party/sglang/python/sglang/srt/mem_cache/session_aware_cache.py
third_party/sglang/python/sglang/srt/mem_cache/swa_memory_pool.py
third_party/sglang/python/sglang/srt/mem_cache/swa_radix_cache.py
third_party/sglang/python/sglang/srt/mem_cache/utils.py
third_party/sglang/python/sglang/srt/mem_cache/cpp_radix_tree/common.h
third_party/sglang/python/sglang/srt/mem_cache/cpp_radix_tree/radix_tree.py
third_party/sglang/python/sglang/srt/mem_cache/cpp_radix_tree/tree_v2.cpp
third_party/sglang/python/sglang/srt/mem_cache/cpp_radix_tree/tree_v2.h
third_party/sglang/python/sglang/srt/mem_cache/cpp_radix_tree/tree_v2_binding.cpp
third_party/sglang/python/sglang/srt/mem_cache/cpp_radix_tree/tree_v2_debug.cpp
third_party/sglang/python/sglang/srt/mem_cache/cpp_radix_tree/tree_v2_impl.h
third_party/sglang/python/sglang/srt/mem_cache/cpp_radix_tree/tree_v2_node.h
third_party/sglang/python/sglang/srt/mem_cache/hybrid_cache/hybrid_cache_controller.py
third_party/sglang/python/sglang/srt/mem_cache/sparsity/__init__.py
third_party/sglang/python/sglang/srt/mem_cache/sparsity/factory.py
third_party/sglang/python/sglang/srt/mem_cache/sparsity/algorithms/__init__.py
third_party/sglang/python/sglang/srt/mem_cache/sparsity/algorithms/base_algorithm.py
third_party/sglang/python/sglang/srt/mem_cache/sparsity/algorithms/deepseek_nsa.py
third_party/sglang/python/sglang/srt/mem_cache/sparsity/algorithms/quest_algorithm.py
third_party/sglang/python/sglang/srt/mem_cache/sparsity/backend/__init__.py
third_party/sglang/python/sglang/srt/mem_cache/sparsity/backend/backend_adaptor.py
third_party/sglang/python/sglang/srt/mem_cache/sparsity/core/__init__.py
third_party/sglang/python/sglang/srt/mem_cache/sparsity/core/sparse_coordinator.py
third_party/sglang/python/sglang/srt/mem_cache/storage/__init__.py
third_party/sglang/python/sglang/srt/mem_cache/storage/backend_factory.py
third_party/sglang/python/sglang/srt/mem_cache/storage/aibrix_kvcache/README.md
third_party/sglang/python/sglang/srt/mem_cache/storage/aibrix_kvcache/aibrix_kvcache_storage.py
third_party/sglang/python/sglang/srt/mem_cache/storage/aibrix_kvcache/unit_test.py
third_party/sglang/python/sglang/srt/mem_cache/storage/eic/README.md
third_party/sglang/python/sglang/srt/mem_cache/storage/eic/eic_storage.py
third_party/sglang/python/sglang/srt/mem_cache/storage/eic/test_unit.py
third_party/sglang/python/sglang/srt/mem_cache/storage/hf3fs/hf3fs_client.py
third_party/sglang/python/sglang/srt/mem_cache/storage/hf3fs/hf3fs_usrbio_client.py
third_party/sglang/python/sglang/srt/mem_cache/storage/hf3fs/hf3fs_utils.cpp
third_party/sglang/python/sglang/srt/mem_cache/storage/hf3fs/mini_3fs_metadata_server.py
third_party/sglang/python/sglang/srt/mem_cache/storage/hf3fs/storage_hf3fs.py
third_party/sglang/python/sglang/srt/mem_cache/storage/hf3fs/test_hf3fs_utils.py
third_party/sglang/python/sglang/srt/mem_cache/storage/hf3fs/docs/README.md
third_party/sglang/python/sglang/srt/mem_cache/storage/hf3fs/docs/deploy_sglang_3fs_multinode.md
third_party/sglang/python/sglang/srt/mem_cache/storage/hf3fs/docs/setup_usrbio_client.md
third_party/sglang/python/sglang/srt/mem_cache/storage/lmcache/README.md
third_party/sglang/python/sglang/srt/mem_cache/storage/lmcache/example_config.yaml
third_party/sglang/python/sglang/srt/mem_cache/storage/lmcache/lmc_radix_cache.py
third_party/sglang/python/sglang/srt/mem_cache/storage/lmcache/unit_test.py
third_party/sglang/python/sglang/srt/mem_cache/storage/mooncake_store/README.md
third_party/sglang/python/sglang/srt/mem_cache/storage/mooncake_store/embedding_cache_controller.py
third_party/sglang/python/sglang/srt/mem_cache/storage/mooncake_store/mooncake_embedding_store.py
third_party/sglang/python/sglang/srt/mem_cache/storage/mooncake_store/mooncake_store.py
third_party/sglang/python/sglang/srt/mem_cache/storage/mooncake_store/test_mooncake_store.py
third_party/sglang/python/sglang/srt/mem_cache/storage/nixl/README.md
third_party/sglang/python/sglang/srt/mem_cache/storage/nixl/hicache_nixl.py
third_party/sglang/python/sglang/srt/mem_cache/storage/nixl/nixl_utils.py
third_party/sglang/python/sglang/srt/mem_cache/storage/nixl/test_hicache_nixl_storage.py
third_party/sglang/python/sglang/srt/model_executor/cpu_graph_runner.py
third_party/sglang/python/sglang/srt/model_executor/cuda_graph_runner.py
third_party/sglang/python/sglang/srt/model_executor/forward_batch_deepseek_mha_mixin.py
third_party/sglang/python/sglang/srt/model_executor/forward_batch_info.py
third_party/sglang/python/sglang/srt/model_executor/hook_manager.py
third_party/sglang/python/sglang/srt/model_executor/input_buffers.py
third_party/sglang/python/sglang/srt/model_executor/mindspore_runner.py
third_party/sglang/python/sglang/srt/model_executor/model_runner.py
third_party/sglang/python/sglang/srt/model_executor/model_runner_kv_cache_mixin.py
third_party/sglang/python/sglang/srt/model_executor/piecewise_cuda_graph_runner.py
third_party/sglang/python/sglang/srt/model_loader/__init__.py
third_party/sglang/python/sglang/srt/model_loader/ci_weight_validation.py
third_party/sglang/python/sglang/srt/model_loader/loader.py
third_party/sglang/python/sglang/srt/model_loader/parameter_mapper.py
third_party/sglang/python/sglang/srt/model_loader/remote_instance_weight_loader_utils.py
third_party/sglang/python/sglang/srt/model_loader/utils.py
third_party/sglang/python/sglang/srt/model_loader/weight_utils.py
third_party/sglang/python/sglang/srt/models/afmoe.py
third_party/sglang/python/sglang/srt/models/apertus.py
third_party/sglang/python/sglang/srt/models/arcee.py
third_party/sglang/python/sglang/srt/models/baichuan.py
third_party/sglang/python/sglang/srt/models/bailing_moe.py
third_party/sglang/python/sglang/srt/models/bailing_moe_linear.py
third_party/sglang/python/sglang/srt/models/bailing_moe_nextn.py
third_party/sglang/python/sglang/srt/models/bert.py
third_party/sglang/python/sglang/srt/models/chatglm.py
third_party/sglang/python/sglang/srt/models/clip.py
third_party/sglang/python/sglang/srt/models/commandr.py
third_party/sglang/python/sglang/srt/models/dbrx.py
third_party/sglang/python/sglang/srt/models/deepseek.py
third_party/sglang/python/sglang/srt/models/deepseek_janus_pro.py
third_party/sglang/python/sglang/srt/models/deepseek_nextn.py
third_party/sglang/python/sglang/srt/models/deepseek_ocr.py
third_party/sglang/python/sglang/srt/models/deepseek_v2.py
third_party/sglang/python/sglang/srt/models/deepseek_vl2.py
third_party/sglang/python/sglang/srt/models/dots_ocr.py
third_party/sglang/python/sglang/srt/models/dots_vlm.py
third_party/sglang/python/sglang/srt/models/dots_vlm_vit.py
third_party/sglang/python/sglang/srt/models/ernie4.py
third_party/sglang/python/sglang/srt/models/ernie45_moe_vl.py
third_party/sglang/python/sglang/srt/models/ernie45_vl.py
third_party/sglang/python/sglang/srt/models/ernie4_eagle.py
third_party/sglang/python/sglang/srt/models/exaone.py
third_party/sglang/python/sglang/srt/models/exaone4.py
third_party/sglang/python/sglang/srt/models/exaone_moe.py
third_party/sglang/python/sglang/srt/models/exaone_moe_mtp.py
third_party/sglang/python/sglang/srt/models/falcon_h1.py
third_party/sglang/python/sglang/srt/models/gemma.py
third_party/sglang/python/sglang/srt/models/gemma2.py
third_party/sglang/python/sglang/srt/models/gemma2_reward.py
third_party/sglang/python/sglang/srt/models/gemma3_causal.py
third_party/sglang/python/sglang/srt/models/gemma3_mm.py
third_party/sglang/python/sglang/srt/models/gemma3n_audio.py
third_party/sglang/python/sglang/srt/models/gemma3n_causal.py
third_party/sglang/python/sglang/srt/models/gemma3n_mm.py
third_party/sglang/python/sglang/srt/models/glm4.py
third_party/sglang/python/sglang/srt/models/glm4_moe.py
third_party/sglang/python/sglang/srt/models/glm4_moe_lite.py
third_party/sglang/python/sglang/srt/models/glm4_moe_nextn.py
third_party/sglang/python/sglang/srt/models/glm4v.py
third_party/sglang/python/sglang/srt/models/glm4v_moe.py
third_party/sglang/python/sglang/srt/models/glm_ocr.py
third_party/sglang/python/sglang/srt/models/glm_ocr_nextn.py
third_party/sglang/python/sglang/srt/models/glmasr.py
third_party/sglang/python/sglang/srt/models/gpt2.py
third_party/sglang/python/sglang/srt/models/gpt_bigcode.py
third_party/sglang/python/sglang/srt/models/gpt_j.py
third_party/sglang/python/sglang/srt/models/gpt_oss.py
third_party/sglang/python/sglang/srt/models/granite.py
third_party/sglang/python/sglang/srt/models/granitemoe.py
third_party/sglang/python/sglang/srt/models/granitemoehybrid.py
third_party/sglang/python/sglang/srt/models/grok.py
third_party/sglang/python/sglang/srt/models/hunyuan.py
third_party/sglang/python/sglang/srt/models/idefics2.py
third_party/sglang/python/sglang/srt/models/internlm2.py
third_party/sglang/python/sglang/srt/models/internlm2_reward.py
third_party/sglang/python/sglang/srt/models/interns1.py
third_party/sglang/python/sglang/srt/models/interns1pro.py
third_party/sglang/python/sglang/srt/models/internvl.py
third_party/sglang/python/sglang/srt/models/iquest_loopcoder.py
third_party/sglang/python/sglang/srt/models/jet_nemotron.py
third_party/sglang/python/sglang/srt/models/jet_vlm.py
third_party/sglang/python/sglang/srt/models/kimi_k25.py
third_party/sglang/python/sglang/srt/models/kimi_linear.py
third_party/sglang/python/sglang/srt/models/kimi_vl.py
third_party/sglang/python/sglang/srt/models/kimi_vl_moonvit.py
third_party/sglang/python/sglang/srt/models/lfm2.py
third_party/sglang/python/sglang/srt/models/lfm2_moe.py
third_party/sglang/python/sglang/srt/models/lfm2_vl.py
third_party/sglang/python/sglang/srt/models/lightonocr.py
third_party/sglang/python/sglang/srt/models/llada2.py
third_party/sglang/python/sglang/srt/models/llama.py
third_party/sglang/python/sglang/srt/models/llama4.py
third_party/sglang/python/sglang/srt/models/llama_classification.py
third_party/sglang/python/sglang/srt/models/llama_eagle.py
third_party/sglang/python/sglang/srt/models/llama_eagle3.py
third_party/sglang/python/sglang/srt/models/llama_embedding.py
third_party/sglang/python/sglang/srt/models/llama_reward.py
third_party/sglang/python/sglang/srt/models/llava.py
third_party/sglang/python/sglang/srt/models/llavavid.py
third_party/sglang/python/sglang/srt/models/longcat_flash.py
third_party/sglang/python/sglang/srt/models/longcat_flash_nextn.py
third_party/sglang/python/sglang/srt/models/midashenglm.py
third_party/sglang/python/sglang/srt/models/mimo.py
third_party/sglang/python/sglang/srt/models/mimo_mtp.py
third_party/sglang/python/sglang/srt/models/mimo_v2_flash.py
third_party/sglang/python/sglang/srt/models/mimo_v2_flash_nextn.py
third_party/sglang/python/sglang/srt/models/mindspore.py
third_party/sglang/python/sglang/srt/models/minicpm.py
third_party/sglang/python/sglang/srt/models/minicpm3.py
third_party/sglang/python/sglang/srt/models/minicpmo.py
third_party/sglang/python/sglang/srt/models/minicpmv.py
third_party/sglang/python/sglang/srt/models/minimax_m2.py
third_party/sglang/python/sglang/srt/models/ministral3.py
third_party/sglang/python/sglang/srt/models/mistral.py
third_party/sglang/python/sglang/srt/models/mistral_large_3.py
third_party/sglang/python/sglang/srt/models/mistral_large_3_eagle.py
third_party/sglang/python/sglang/srt/models/mixtral.py
third_party/sglang/python/sglang/srt/models/mixtral_quant.py
third_party/sglang/python/sglang/srt/models/mllama.py
third_party/sglang/python/sglang/srt/models/mllama4.py
third_party/sglang/python/sglang/srt/models/nano_nemotron_vl.py
third_party/sglang/python/sglang/srt/models/nemotron_h.py
third_party/sglang/python/sglang/srt/models/nemotron_h_mtp.py
third_party/sglang/python/sglang/srt/models/nemotron_nas.py
third_party/sglang/python/sglang/srt/models/nvila.py
third_party/sglang/python/sglang/srt/models/nvila_lite.py
third_party/sglang/python/sglang/srt/models/olmo.py
third_party/sglang/python/sglang/srt/models/olmo2.py
third_party/sglang/python/sglang/srt/models/olmoe.py
third_party/sglang/python/sglang/srt/models/opt.py
third_party/sglang/python/sglang/srt/models/orion.py
third_party/sglang/python/sglang/srt/models/paddleocr_vl.py
third_party/sglang/python/sglang/srt/models/persimmon.py
third_party/sglang/python/sglang/srt/models/phi.py
third_party/sglang/python/sglang/srt/models/phi3_small.py
third_party/sglang/python/sglang/srt/models/phi4mm.py
third_party/sglang/python/sglang/srt/models/phi4mm_audio.py
third_party/sglang/python/sglang/srt/models/phi4mm_utils.py
third_party/sglang/python/sglang/srt/models/phimoe.py
third_party/sglang/python/sglang/srt/models/pixtral.py
third_party/sglang/python/sglang/srt/models/points_v15_chat.py
third_party/sglang/python/sglang/srt/models/qwen.py
third_party/sglang/python/sglang/srt/models/qwen2.py
third_party/sglang/python/sglang/srt/models/qwen2_5_vl.py
third_party/sglang/python/sglang/srt/models/qwen2_audio.py
third_party/sglang/python/sglang/srt/models/qwen2_classification.py
third_party/sglang/python/sglang/srt/models/qwen2_eagle.py
third_party/sglang/python/sglang/srt/models/qwen2_moe.py
third_party/sglang/python/sglang/srt/models/qwen2_rm.py
third_party/sglang/python/sglang/srt/models/qwen2_vl.py
third_party/sglang/python/sglang/srt/models/qwen3.py
third_party/sglang/python/sglang/srt/models/qwen3_5.py
third_party/sglang/python/sglang/srt/models/qwen3_5_mtp.py
third_party/sglang/python/sglang/srt/models/qwen3_classification.py
third_party/sglang/python/sglang/srt/models/qwen3_moe.py
third_party/sglang/python/sglang/srt/models/qwen3_next.py
third_party/sglang/python/sglang/srt/models/qwen3_next_mtp.py
third_party/sglang/python/sglang/srt/models/qwen3_omni_moe.py
third_party/sglang/python/sglang/srt/models/qwen3_rm.py
third_party/sglang/python/sglang/srt/models/qwen3_vl.py
third_party/sglang/python/sglang/srt/models/qwen3_vl_moe.py
third_party/sglang/python/sglang/srt/models/radio.py
third_party/sglang/python/sglang/srt/models/registry.py
third_party/sglang/python/sglang/srt/models/roberta.py
third_party/sglang/python/sglang/srt/models/sarashina2_vision.py
third_party/sglang/python/sglang/srt/models/sarvam_moe.py
third_party/sglang/python/sglang/srt/models/sdar.py
third_party/sglang/python/sglang/srt/models/sdar_moe.py
third_party/sglang/python/sglang/srt/models/siglip.py
third_party/sglang/python/sglang/srt/models/siglip2.py
third_party/sglang/python/sglang/srt/models/solar.py
third_party/sglang/python/sglang/srt/models/stablelm.py
third_party/sglang/python/sglang/srt/models/starcoder2.py
third_party/sglang/python/sglang/srt/models/step3_vl.py
third_party/sglang/python/sglang/srt/models/step3_vl_10b.py
third_party/sglang/python/sglang/srt/models/step3p5.py
third_party/sglang/python/sglang/srt/models/step3p5_mtp.py
third_party/sglang/python/sglang/srt/models/teleflm.py
third_party/sglang/python/sglang/srt/models/torch_native_llama.py
third_party/sglang/python/sglang/srt/models/transformers.py
third_party/sglang/python/sglang/srt/models/utils.py
third_party/sglang/python/sglang/srt/models/whisper.py
third_party/sglang/python/sglang/srt/models/xverse.py
third_party/sglang/python/sglang/srt/models/xverse_moe.py
third_party/sglang/python/sglang/srt/models/yivl.py
third_party/sglang/python/sglang/srt/models/deepseek_common/__init__.py
third_party/sglang/python/sglang/srt/models/deepseek_common/attention_backend_handler.py
third_party/sglang/python/sglang/srt/models/deepseek_common/deepseek_weight_loader.py
third_party/sglang/python/sglang/srt/models/deepseek_common/utils.py
third_party/sglang/python/sglang/srt/models/deepseek_common/attention_forward_methods/__init__.py
third_party/sglang/python/sglang/srt/models/deepseek_common/attention_forward_methods/forward_methods.py
third_party/sglang/python/sglang/srt/models/deepseek_common/attention_forward_methods/forward_mha.py
third_party/sglang/python/sglang/srt/models/deepseek_common/attention_forward_methods/forward_mla.py
third_party/sglang/python/sglang/srt/models/deepseek_common/attention_forward_methods/forward_mla_fused_rope_cpu.py
third_party/sglang/python/sglang/srt/models/deepseek_common/attention_forward_methods/forward_mla_fused_rope_rocm.py
third_party/sglang/python/sglang/srt/multimodal/customized_mm_processor_utils.py
third_party/sglang/python/sglang/srt/multimodal/internvl_utils.py
third_party/sglang/python/sglang/srt/multimodal/internvl_vit_cuda_graph_runner.py
third_party/sglang/python/sglang/srt/multimodal/mm_utils.py
third_party/sglang/python/sglang/srt/multimodal/vit_cuda_graph_runner.py
third_party/sglang/python/sglang/srt/multimodal/evs/README.md
third_party/sglang/python/sglang/srt/multimodal/evs/__init__.py
third_party/sglang/python/sglang/srt/multimodal/evs/evs_core.py
third_party/sglang/python/sglang/srt/multimodal/evs/evs_module.py
third_party/sglang/python/sglang/srt/multimodal/evs/evs_processor.py
third_party/sglang/python/sglang/srt/multimodal/processors/base_processor.py
third_party/sglang/python/sglang/srt/multimodal/processors/clip.py
third_party/sglang/python/sglang/srt/multimodal/processors/deepseek_ocr.py
third_party/sglang/python/sglang/srt/multimodal/processors/deepseek_vl_v2.py
third_party/sglang/python/sglang/srt/multimodal/processors/dots_vlm.py
third_party/sglang/python/sglang/srt/multimodal/processors/ernie45_vl.py
third_party/sglang/python/sglang/srt/multimodal/processors/gemma3.py
third_party/sglang/python/sglang/srt/multimodal/processors/gemma3n.py
third_party/sglang/python/sglang/srt/multimodal/processors/glm4v.py
third_party/sglang/python/sglang/srt/multimodal/processors/glmasr.py
third_party/sglang/python/sglang/srt/multimodal/processors/interns1pro.py
third_party/sglang/python/sglang/srt/multimodal/processors/internvl.py
third_party/sglang/python/sglang/srt/multimodal/processors/janus_pro.py
third_party/sglang/python/sglang/srt/multimodal/processors/kimi_k25.py
third_party/sglang/python/sglang/srt/multimodal/processors/kimi_vl.py
third_party/sglang/python/sglang/srt/multimodal/processors/lfm2_vl.py
third_party/sglang/python/sglang/srt/multimodal/processors/lightonocr.py
third_party/sglang/python/sglang/srt/multimodal/processors/llava.py
third_party/sglang/python/sglang/srt/multimodal/processors/midashenglm.py
third_party/sglang/python/sglang/srt/multimodal/processors/minicpm.py
third_party/sglang/python/sglang/srt/multimodal/processors/mlama.py
third_party/sglang/python/sglang/srt/multimodal/processors/mllama4.py
third_party/sglang/python/sglang/srt/multimodal/processors/nano_nemotron_vl.py
third_party/sglang/python/sglang/srt/multimodal/processors/nvila.py
third_party/sglang/python/sglang/srt/multimodal/processors/paddleocr_vlm.py
third_party/sglang/python/sglang/srt/multimodal/processors/phi4mm.py
third_party/sglang/python/sglang/srt/multimodal/processors/pixtral.py
third_party/sglang/python/sglang/srt/multimodal/processors/points_v15_chat.py
third_party/sglang/python/sglang/srt/multimodal/processors/qwen_audio.py
third_party/sglang/python/sglang/srt/multimodal/processors/qwen_vl.py
third_party/sglang/python/sglang/srt/multimodal/processors/sarashina2_vision.py
third_party/sglang/python/sglang/srt/multimodal/processors/step3_vl.py
third_party/sglang/python/sglang/srt/multimodal/processors/transformers_auto.py
third_party/sglang/python/sglang/srt/multimodal/processors/whisper.py
third_party/sglang/python/sglang/srt/multiplex/multiplexing_mixin.py
third_party/sglang/python/sglang/srt/multiplex/pdmux_context.py
third_party/sglang/python/sglang/srt/observability/cpu_monitor.py
third_party/sglang/python/sglang/srt/observability/func_timer.py
third_party/sglang/python/sglang/srt/observability/label_transform.py
third_party/sglang/python/sglang/srt/observability/metrics_collector.py
third_party/sglang/python/sglang/srt/observability/req_time_stats.py
third_party/sglang/python/sglang/srt/observability/request_metrics_exporter.py
third_party/sglang/python/sglang/srt/observability/scheduler_metrics_mixin.py
third_party/sglang/python/sglang/srt/observability/startup_func_log_and_timer.py
third_party/sglang/python/sglang/srt/observability/trace.py
third_party/sglang/python/sglang/srt/observability/utils.py
third_party/sglang/python/sglang/srt/parser/code_completion_parser.py
third_party/sglang/python/sglang/srt/parser/conversation.py
third_party/sglang/python/sglang/srt/parser/harmony_parser.py
third_party/sglang/python/sglang/srt/parser/jinja_template_utils.py
third_party/sglang/python/sglang/srt/parser/reasoning_parser.py
third_party/sglang/python/sglang/srt/ray/__init__.py
third_party/sglang/python/sglang/srt/ray/engine.py
third_party/sglang/python/sglang/srt/ray/http_server.py
third_party/sglang/python/sglang/srt/ray/scheduler_actor.py
third_party/sglang/python/sglang/srt/sampling/custom_logit_processor.py
third_party/sglang/python/sglang/srt/sampling/sampling_batch_info.py
third_party/sglang/python/sglang/srt/sampling/sampling_params.py
third_party/sglang/python/sglang/srt/sampling/penaltylib/__init__.py
third_party/sglang/python/sglang/srt/sampling/penaltylib/frequency_penalty.py
third_party/sglang/python/sglang/srt/sampling/penaltylib/min_new_tokens.py
third_party/sglang/python/sglang/srt/sampling/penaltylib/orchestrator.py
third_party/sglang/python/sglang/srt/sampling/penaltylib/presence_penalty.py
third_party/sglang/python/sglang/srt/sampling/penaltylib/repetition_penalty.py
third_party/sglang/python/sglang/srt/speculative/base_spec_worker.py
third_party/sglang/python/sglang/srt/speculative/draft_utils.py
third_party/sglang/python/sglang/srt/speculative/eagle_draft_cuda_graph_runner.py
third_party/sglang/python/sglang/srt/speculative/eagle_draft_extend_cuda_graph_runner.py
third_party/sglang/python/sglang/srt/speculative/eagle_info.py
third_party/sglang/python/sglang/srt/speculative/eagle_info_v2.py
third_party/sglang/python/sglang/srt/speculative/eagle_utils.py
third_party/sglang/python/sglang/srt/speculative/eagle_worker.py
third_party/sglang/python/sglang/srt/speculative/eagle_worker_v2.py
third_party/sglang/python/sglang/srt/speculative/multi_layer_eagle_draft_extend_cuda_graph_runner.py
third_party/sglang/python/sglang/srt/speculative/multi_layer_eagle_utils.py
third_party/sglang/python/sglang/srt/speculative/multi_layer_eagle_worker.py
third_party/sglang/python/sglang/srt/speculative/multi_layer_eagle_worker_v2.py
third_party/sglang/python/sglang/srt/speculative/ngram_info.py
third_party/sglang/python/sglang/srt/speculative/ngram_worker.py
third_party/sglang/python/sglang/srt/speculative/spec_info.py
third_party/sglang/python/sglang/srt/speculative/spec_utils.py
third_party/sglang/python/sglang/srt/speculative/standalone_worker.py
third_party/sglang/python/sglang/srt/speculative/standalone_worker_v2.py
third_party/sglang/python/sglang/srt/speculative/cpp_ngram/ngram_corpus.py
third_party/sglang/python/sglang/srt/tokenizer/tiktoken_tokenizer.py
third_party/sglang/python/sglang/srt/tp_invariant_ops/__init__.py
third_party/sglang/python/sglang/srt/tp_invariant_ops/tp_invariant_ops.py
third_party/sglang/python/sglang/srt/true_on_policy/__init__.py
third_party/sglang/python/sglang/srt/true_on_policy/config.py
third_party/sglang/python/sglang/srt/true_on_policy/contracts.py
third_party/sglang/python/sglang/srt/true_on_policy/schema.py
third_party/sglang/python/sglang/srt/utils/__init__.py
third_party/sglang/python/sglang/srt/utils/aio_rwlock.py
third_party/sglang/python/sglang/srt/utils/auth.py
third_party/sglang/python/sglang/srt/utils/bench_utils.py
third_party/sglang/python/sglang/srt/utils/common.py
third_party/sglang/python/sglang/srt/utils/cuda_ipc_transport_utils.py
third_party/sglang/python/sglang/srt/utils/custom_op.py
third_party/sglang/python/sglang/srt/utils/device_timer.py
third_party/sglang/python/sglang/srt/utils/gauge_histogram.py
third_party/sglang/python/sglang/srt/utils/hf_transformers_utils.py
third_party/sglang/python/sglang/srt/utils/host_shared_memory.py
third_party/sglang/python/sglang/srt/utils/http_middleware_patch.py
third_party/sglang/python/sglang/srt/utils/json_response.py
third_party/sglang/python/sglang/srt/utils/log_utils.py
third_party/sglang/python/sglang/srt/utils/mistral_utils.py
third_party/sglang/python/sglang/srt/utils/model_file_verifier.py
third_party/sglang/python/sglang/srt/utils/multi_stream_utils.py
third_party/sglang/python/sglang/srt/utils/network.py
third_party/sglang/python/sglang/srt/utils/numa_utils.py
third_party/sglang/python/sglang/srt/utils/nvtx_pytorch_hooks.py
third_party/sglang/python/sglang/srt/utils/offloader.py
third_party/sglang/python/sglang/srt/utils/patch_tokenizer.py
third_party/sglang/python/sglang/srt/utils/patch_torch.py
third_party/sglang/python/sglang/srt/utils/poll_based_barrier.py
third_party/sglang/python/sglang/srt/utils/profile_merger.py
third_party/sglang/python/sglang/srt/utils/profile_utils.py
third_party/sglang/python/sglang/srt/utils/request_logger.py
third_party/sglang/python/sglang/srt/utils/rpd_utils.py
third_party/sglang/python/sglang/srt/utils/runai_utils.py
third_party/sglang/python/sglang/srt/utils/scheduler_status_logger.py
third_party/sglang/python/sglang/srt/utils/slow_rank_detector.py
third_party/sglang/python/sglang/srt/utils/tensor_bridge.py
third_party/sglang/python/sglang/srt/utils/torch_memory_saver_adapter.py
third_party/sglang/python/sglang/srt/utils/video_decoder.py
third_party/sglang/python/sglang/srt/utils/watchdog.py
third_party/sglang/python/sglang/srt/utils/weight_checker.py
third_party/sglang/python/sglang/srt/weight_sync/tensor_bucket.py
third_party/sglang/python/sglang/srt/weight_sync/utils.py